diff --git a/include/mimalloc/internal.h b/include/mimalloc/internal.h
index e316de94..208989e3 100644
--- a/include/mimalloc/internal.h
+++ b/include/mimalloc/internal.h
@@ -147,6 +147,7 @@ mi_arena_t* _mi_arena_from_id(mi_arena_id_t id);
 
 void*       _mi_arena_alloc(mi_subproc_t* subproc, size_t size, bool commit, bool allow_large, mi_arena_t* req_arena, size_t tseq, mi_memid_t* memid);
 void*       _mi_arena_alloc_aligned(mi_subproc_t* subproc, size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_arena_t* req_arena, size_t tseq, mi_memid_t* memid);
+void        _mi_arena_free(void* p, size_t size, mi_memid_t memid);
 bool        _mi_arena_memid_is_suitable(mi_memid_t memid, mi_arena_t* request_arena);
 bool        _mi_arena_contains(const void* p);
 void        _mi_arenas_collect(bool force_purge);
@@ -421,11 +422,11 @@ static inline bool mi_heap_is_initialized(mi_heap_t* heap) {
   return (heap != &_mi_heap_empty);
 }
 
-static inline uintptr_t _mi_ptr_cookie(const void* p) {
-  extern mi_heap_t _mi_heap_main;
-  mi_assert_internal(_mi_heap_main.cookie != 0);
-  return ((uintptr_t)p ^ _mi_heap_main.cookie);
-}
+//static inline uintptr_t _mi_ptr_cookie(const void* p) {
+//  extern mi_heap_t _mi_heap_main;
+//  mi_assert_internal(_mi_heap_main.cookie != 0);
+//  return ((uintptr_t)p ^ _mi_heap_main.cookie);
+//}
 
 
 /* -----------------------------------------------------------
diff --git a/include/mimalloc/types.h b/include/mimalloc/types.h
index 59393848..461b5393 100644
--- a/include/mimalloc/types.h
+++ b/include/mimalloc/types.h
@@ -396,7 +396,6 @@ struct mi_heap_s {
   mi_tld_t*             tld;                                 // thread-local data
   mi_arena_t*           exclusive_arena;                     // if the heap should only allocate from a specific arena (or NULL)
   uintptr_t             cookie;                              // random cookie to verify pointers (see `_mi_ptr_cookie`)
-  uintptr_t             keys[2];                             // two random keys used to encode the `thread_delayed_free` list
   mi_random_ctx_t       random;                              // random number context used for secure allocation
   size_t                page_count;                          // total number of pages in the `pages` queues.
   size_t                page_retired_min;                    // smallest retired index (retired pages are fully free, but still in the page queues)
@@ -522,21 +521,13 @@ void __mi_stat_counter_increase_mt(mi_stat_counter_t* stat, size_t amount);
 #define mi_os_stat_increase(stat,amount)                        mi_subproc_stat_increase(_mi_subproc(),stat,amount)
 #define mi_os_stat_decrease(stat,amount)                        mi_subproc_stat_decrease(_mi_subproc(),stat,amount)
 
-#define mi_tld_stat_counter_increase(tld,stat,amount)           __mi_stat_counter_increase( &(tld)->stats.stat, amount)
-#define mi_tld_stat_increase(tld,stat,amount)                   __mi_stat_increase( &(tld)->stats.stat, amount)
-#define mi_tld_stat_decrease(tld,stat,amount)                   __mi_stat_decrease( &(tld)->stats.stat, amount)
+#define mi_heap_stat_counter_increase(heap,stat,amount)         __mi_stat_counter_increase( &(heap)->tld->stats.stat, amount)
+#define mi_heap_stat_increase(heap,stat,amount)                 __mi_stat_increase( &(heap)->tld->stats.stat, amount)
+#define mi_heap_stat_decrease(heap,stat,amount)                 __mi_stat_decrease( &(heap)->tld->stats.stat, amount)
 
-#define mi_debug_tld_stat_counter_increase(tld,stat,amount)     mi_debug_stat_counter_increase( (tld)->stats.stat, amount)
-#define mi_debug_tld_stat_increase(tld,stat,amount)             mi_debug_stat_increase( (tld)->stats.stat, amount)
-#define mi_debug_tld_stat_decrease(tld,stat,amount)             mi_debug_stat_decrease( (tld)->stats.stat, amount)
-
-#define mi_heap_stat_counter_increase(heap,stat,amount)         mi_tld_stat_counter_increase((heap)->tld, stat, amount)
-#define mi_heap_stat_increase(heap,stat,amount)                 mi_tld_stat_increase( (heap)->tld, stat, amount)
-#define mi_heap_stat_decrease(heap,stat,amount)                 mi_tld_stat_decrease( (heap)->tld, stat, amount)
-
-#define mi_debug_heap_stat_counter_increase(heap,stat,amount)   mi_debug_tld_stat_counter_increase((heap)->tld, stat, amount)
-#define mi_debug_heap_stat_increase(heap,stat,amount)           mi_debug_tld_stat_increase( (heap)->tld, stat, amount)
-#define mi_debug_heap_stat_decrease(heap,stat,amount)           mi_debug_tld_stat_decrease( (heap)->tld, stat, amount)
+#define mi_debug_heap_stat_counter_increase(heap,stat,amount)   mi_debug_stat_counter_increase( (heap)->tld->stats.stat, amount)
+#define mi_debug_heap_stat_increase(heap,stat,amount)           mi_debug_stat_increase( (heap)->tld->stats.stat, amount)
+#define mi_debug_heap_stat_decrease(heap,stat,amount)           mi_debug_stat_decrease( (heap)->tld->stats.stat, amount)
 
 
 // ------------------------------------------------------
diff --git a/src/arena-meta.c b/src/arena-meta.c
index a5dc8e75..065a1331 100644
--- a/src/arena-meta.c
+++ b/src/arena-meta.c
@@ -148,11 +148,8 @@ mi_decl_noinline void _mi_meta_free(void* p, size_t size, mi_memid_t memid) {
     _mi_memzero_aligned(mi_meta_block_start(mpage, block_idx), block_count*MI_META_BLOCK_SIZE);
     mi_bitmap_setN(&mpage->blocks_free, block_idx, block_count,NULL);
   }
-  else if (mi_memid_is_os(memid)) {
-    _mi_os_free(p, size, memid);    
-  }
   else {
-    mi_assert_internal(mi_memid_needs_no_free(memid));
+    _mi_arena_free(p,size,memid);
   }
 }
 
diff --git a/src/arena.c b/src/arena.c
index c4b02cf6..869cba49 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -762,8 +762,6 @@ mi_page_t* _mi_arena_page_alloc(mi_heap_t* heap, size_t block_size, size_t block
   return page;
 }
 
-static void mi_arena_free(void* p, size_t size, mi_memid_t memid);
-
 void _mi_arena_page_free(mi_page_t* page) {
   mi_assert_internal(_mi_is_aligned(page, MI_PAGE_ALIGN));
   mi_assert_internal(_mi_ptr_page(page)==page);
@@ -794,7 +792,7 @@ void _mi_arena_page_free(mi_page_t* page) {
   if (page->memid.memkind == MI_MEM_ARENA) {
     mi_bitmap_clear(page->memid.mem.arena.arena->pages, page->memid.mem.arena.slice_index);
   }
-  mi_arena_free(page, mi_memid_size(page->memid), page->memid);
+  _mi_arena_free(page, mi_memid_size(page->memid), page->memid);
 }
 
 /* -----------------------------------------------------------
@@ -920,7 +918,7 @@ void _mi_arena_reclaim_all_abandoned(mi_heap_t* heap) {
 static void mi_arena_schedule_purge(mi_arena_t* arena, size_t slice_index, size_t slices);
 static void mi_arenas_try_purge(bool force, bool visit_all);
 
-static void mi_arena_free(void* p, size_t size, mi_memid_t memid) {
+void _mi_arena_free(void* p, size_t size, mi_memid_t memid) {
   if (p==NULL) return;
   if (size==0) return;
 
diff --git a/src/heap.c b/src/heap.c
index d82b383f..f47aaad9 100644
--- a/src/heap.c
+++ b/src/heap.c
@@ -213,8 +213,8 @@ void _mi_heap_init(mi_heap_t* heap, mi_arena_id_t arena_id, bool noreclaim, uint
     _mi_random_split(&heap->tld->heap_backing->random, &heap->random);
   }
   heap->cookie  = _mi_heap_random_next(heap) | 1;
-  heap->keys[0] = _mi_heap_random_next(heap);
-  heap->keys[1] = _mi_heap_random_next(heap);
+  //heap->keys[0] = _mi_heap_random_next(heap);
+  //heap->keys[1] = _mi_heap_random_next(heap);*/
   _mi_heap_guarded_init(heap);
 
   // push on the thread local heaps list
@@ -227,7 +227,15 @@ mi_heap_t* _mi_heap_create(int heap_tag, bool allow_destroy, mi_arena_id_t arena
   mi_assert(heap_tag >= 0 && heap_tag < 256);
   // allocate and initialize a heap
   mi_memid_t memid;
-  mi_heap_t* heap = (mi_heap_t*)_mi_meta_zalloc(sizeof(mi_heap_t), &memid);
+  mi_heap_t* heap; 
+  if (arena_id == _mi_arena_id_none()) {
+    heap = (mi_heap_t*)_mi_meta_zalloc(sizeof(mi_heap_t), &memid);
+  }
+  else {
+    // heaps associated wita a specific arena are allocated in that arena
+    // note: takes up at least one slice which is quite wasteful...
+    heap = (mi_heap_t*)_mi_arena_alloc(_mi_subproc(), sizeof(mi_heap_t), true, true, _mi_arena_from_id(arena_id), tld->thread_seq, &memid);
+  }
   if (heap==NULL) {
     _mi_error_message(ENOMEM, "unable to allocate heap meta-data\n");
     return NULL;
diff --git a/src/init.c b/src/init.c
index 1968ef68..2f147e55 100644
--- a/src/init.c
+++ b/src/init.c
@@ -115,7 +115,7 @@ mi_decl_cache_align const mi_heap_t _mi_heap_empty = {
   &tld_empty,             // tld
   NULL,                   // exclusive_arena
   0,                      // cookie
-  { 0, 0 },               // keys
+  //{ 0, 0 },               // keys
   { {0}, {0}, 0, true },  // random
   0,                      // page count
   MI_BIN_FULL, 0,         // page retired min/max
@@ -149,9 +149,9 @@ static mi_decl_cache_align mi_tld_t tld_main = {
 
 mi_decl_cache_align mi_heap_t heap_main = {
   &tld_main,              // thread local data
+  NULL,                   // exclusive arena
   0,                      // initial cookie
-  0,                      // arena id
-  { 0, 0 },               // the key of the main heap can be fixed (unlike page keys that need to be secure!)
+  //{ 0, 0 },               // the key of the main heap can be fixed (unlike page keys that need to be secure!)
   { {0x846ca68b}, {0}, 0, true },  // random
   0,                      // page count
   MI_BIN_FULL, 0,         // page retired min/max
@@ -248,8 +248,8 @@ static void mi_heap_main_init(void) {
       _mi_random_init(&heap_main.random);
     #endif
     heap_main.cookie  = _mi_heap_random_next(&heap_main);
-    heap_main.keys[0] = _mi_heap_random_next(&heap_main);
-    heap_main.keys[1] = _mi_heap_random_next(&heap_main);
+    //heap_main.keys[0] = _mi_heap_random_next(&heap_main);
+    //heap_main.keys[1] = _mi_heap_random_next(&heap_main);
     _mi_heap_guarded_init(&heap_main);
     heap_main.allow_page_abandon = (mi_option_get(mi_option_full_page_retain) >= 0);
     heap_main.full_page_retain   = mi_option_get_clamp(mi_option_full_page_retain, -1, 32);