diff --git a/include/mimalloc/internal.h b/include/mimalloc/internal.h
index 01b7076b..ec106047 100644
--- a/include/mimalloc/internal.h
+++ b/include/mimalloc/internal.h
@@ -598,7 +598,7 @@ static inline bool mi_page_mostly_used(const mi_page_t* page) {
 
 static inline bool mi_page_is_abandoned(const mi_page_t* page) {
   // note: the xheap field of an abandoned heap is set to the subproc (for fast reclaim-on-free)
-  return (mi_page_thread_id(page) == 0);
+  return (mi_atomic_load_acquire(&page->xthread_id) == 0);
 }
 
 static inline bool mi_page_is_huge(const mi_page_t* page) {
diff --git a/src/arena.c b/src/arena.c
index 66f83d4f..a713a110 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -646,11 +646,12 @@ bool _mi_arena_try_reclaim(mi_heap_t* heap, mi_page_t* page) {
   // we only call this when on free (and thus there is still an object alive in the page)
   mi_memid_t memid = page->memid;
   if (!_mi_arena_memid_is_suitable(memid, heap->arena_id)) return false;  // don't reclaim between exclusive and non-exclusive arena's
+  if (mi_atomic_load_acquire(&page->xheap) != (uintptr_t)heap->tld->subproc) return false;
 
   if mi_likely(memid.memkind == MI_MEM_ARENA) {
     size_t slice_index;
     mi_arena_t* arena = mi_page_arena(page, &slice_index, NULL);
-    if (arena->subproc != heap->tld->subproc) return false;  // only reclaim within the same subprocess
+    //if (arena->subproc != heap->tld->subproc) return false;  // only reclaim within the same subprocess
 
     // don't reclaim more from a `free` call than half the current segments
     // this is to prevent a pure free-ing thread to start owning too many segments
@@ -665,6 +666,11 @@ bool _mi_arena_try_reclaim(mi_heap_t* heap, mi_page_t* page) {
       mi_assert_internal(!mi_page_is_abandoned(page));
       return true;
     }
+    else {
+      if (mi_page_is_abandoned(page)) {
+        mi_assert(false);
+      }
+    }
   }
   else {
     // A page in OS or external memory
@@ -1089,15 +1095,6 @@ int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserv
 
 
 
-/* -----------------------------------------------------------
-  Abandoned pages
------------------------------------------------------------ */
-
-void mi_arena_page_abandon(mi_page_t* page) {
-  mi_assert_internal(mi_page_is_abandoned(page));
-  if (mi_page_is_full(page)) {}
-}
-
 
 
 /* -----------------------------------------------------------
diff --git a/src/free.c b/src/free.c
index c7d92292..f0ce8c22 100644
--- a/src/free.c
+++ b/src/free.c
@@ -225,24 +225,29 @@ static void mi_decl_noinline mi_free_block_delayed_mt( mi_page_t* page, mi_block
 static void mi_decl_noinline mi_free_block_mt(mi_page_t* page, mi_block_t* block)
 {
   // first see if the page was abandoned and if we can reclaim it into our thread
-  if (mi_page_is_abandoned(page) &&
-      (_mi_option_get_fast(mi_option_abandoned_reclaim_on_free) != 0 ||
-       mi_page_is_singleton(page)  // only one block, and we are free-ing it
-      ) &&
-      mi_prim_get_default_heap() != (mi_heap_t*)&_mi_heap_empty) // and we did not already exit this thread (without this check, a fresh heap will be initalized (issue #944))
-  {
-    // the page is abandoned, try to reclaim it into our heap
-    if (_mi_arena_try_reclaim(mi_heap_get_default(), page)) {  // TODO: avoid putting it in the full free queue
-      mi_assert_internal(_mi_thread_id() == mi_page_thread_id(page));      
-      // mi_assert_internal(mi_heap_get_default()->tld->subproc == page->subproc);
-      mi_free(block);  // recursively free as now it will be a local free in our heap
-      return;
-    }
-    else {
-      mi_assert_internal(!mi_page_is_singleton(page)); // we should have succeeded on singleton pages
+  if (mi_page_is_abandoned(page)) {
+    if (_mi_option_get_fast(mi_option_abandoned_reclaim_on_free) != 0 ||
+      mi_page_is_singleton(page)) {  // only one block, and we are free-ing it
+      if (mi_prim_get_default_heap() != (mi_heap_t*)&_mi_heap_empty) // and we did not already exit this thread (without this check, a fresh heap will be initalized (issue #944))
+      {
+        // the page is abandoned, try to reclaim it into our heap
+        if (_mi_arena_try_reclaim(mi_heap_get_default(), page)) {  // TODO: avoid putting it in the full free queue
+          mi_assert_internal(_mi_thread_id() == mi_page_thread_id(page));
+          // mi_assert_internal(mi_heap_get_default()->tld->subproc == page->subproc);
+          mi_free(block);  // recursively free as now it will be a local free in our heap
+          return;
+        }
+        else {
+          if (mi_page_is_abandoned(page)) {
+            mi_assert(false);
+          }
+          mi_assert_internal(!mi_page_is_singleton(page)); // we should have succeeded on singleton pages
+        }
+      }
     }
   }
 
+
   // The padding check may access the non-thread-owned page for the key values.
   // that is safe as these are constant and the page won't be freed (as the block is not freed yet).
   mi_check_padding(page, block);
diff --git a/src/init.c b/src/init.c
index 16130af7..2378b3c8 100644
--- a/src/init.c
+++ b/src/init.c
@@ -396,8 +396,7 @@ void _mi_tld_init(mi_tld_t* tld, mi_heap_t* bheap) {
   tld->heap_backing = bheap;
   tld->heaps = NULL;
   tld->subproc = &mi_subproc_default;
-  tld->tseq = 0;
-  mi_atomic_add_acq_rel(&mi_tcount, 1);
+  tld->tseq = 0; // mi_atomic_add_acq_rel(&mi_tcount, 1);
   tld->os.stats = &tld->stats;
 }
 
diff --git a/src/os.c b/src/os.c
index 931abc7f..0aa0a681 100644
--- a/src/os.c
+++ b/src/os.c
@@ -219,11 +219,12 @@ static void* mi_os_prim_alloc_aligned(size_t size, size_t alignment, bool commit
   if (!(alignment >= _mi_os_page_size() && ((alignment & (alignment - 1)) == 0))) return NULL;
   size = _mi_align_up(size, _mi_os_page_size());
 
-  const bool use_overalloc = (alignment > mi_os_mem_config.alloc_granularity && alignment <= size/8);
+  // try a direct allocation if the alignment is below the default, or if larger than 1/64 fraction of the size (to avoid waste).
+  const bool try_direct_alloc = (alignment <= mi_os_mem_config.alloc_granularity || alignment > size/64);
 
   // try first with a requested alignment hint (this will usually be aligned directly on Win 10+ or BSD)
   void* p = NULL;
-  if (!use_overalloc) {
+  if (try_direct_alloc) {
     p = mi_os_prim_alloc(size, alignment, commit, allow_large, is_large, is_zero, stats);
   }
 
@@ -234,7 +235,7 @@ static void* mi_os_prim_alloc_aligned(size_t size, size_t alignment, bool commit
   else {
     // if not aligned, free it, overallocate, and unmap around it
     #if !MI_TRACK_ASAN
-    if (!use_overalloc) {
+    if (try_direct_alloc) {
       _mi_warning_message("unable to allocate aligned OS memory directly, fall back to over-allocation (size: 0x%zx bytes, address: %p, alignment: 0x%zx, commit: %d)\n", size, p, alignment, commit);
     }
     #endif
diff --git a/test/test-stress.c b/test/test-stress.c
index 76dfe877..9a89744e 100644
--- a/test/test-stress.c
+++ b/test/test-stress.c
@@ -36,7 +36,7 @@ static int ITER    = 400;
 static int THREADS = 8;
 static int SCALE   = 25;
 static int ITER    = 20;
-#elif defined(xMI_GUARDED)     // with debug guard pages reduce parameters to stay within the azure pipeline limits
+#elif defined(MI_GUARDED)     // with debug guard pages reduce parameters to stay within the azure pipeline limits
 static int THREADS = 8;
 static int SCALE   = 10;
 static int ITER    = 10;