What are reasons for "Cannot allocate memory" except of exceeding address space and memory fragmentation?

The problem is that in a 32-bit application on Mac OS X I receive an error

malloc: *** mmap(size=49721344) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug

For the reference error code is in sys/errno.h:

#define ENOMEM   12  /* Cannot allocate memory */

The memory allocation pattern is like this:

1. First is allocated nearly 250MB of memory
2. Allocate 6 blocks of 32MB
3. Then 27 images each handled like this
   1. Allocate 16MB (image bitmap is loaded)
   2. Allocate 32MB, process it, free these 32MB
   3. Again allocate 32MB, process it, free these 32MB
   4. Free 16MB allocated in step 3.1
4. Free 4 blocks allocated in step 2 (2 blocks are still used)
5. Free 250MB block allocated in step 1
6. Allocate blocks of various size, total size doesn't exceed 250MB. And here I receive the mentioned memory allocation error

I've checked that none of these memory blocks is leaked, so I guess used memory at any given time stays below 1GB, which should be accessible on 32-bit system.

The second guess was memory fragmentation. But I've checked that all block in step 3 reuse same addresses. So I touch less than 1GB of memory - memory fragmentation should not be an issue.

Now I am completely lost what can be a reason for not allocating memory. Also everything works OK when I process less than 27 images. Here is part of heap command result before step 6 for 26 images:

Process 1230: 4 zones
Zone DefaultMallocZone_0x273000: Overall size: 175627KB; 29620 nodes malloced for 68559KB (39% of capacity); largest unused: [0x6f800000-8191KB]
Zone DispatchContinuations_0x292000: Overall size: 4096KB; 1 nodes malloced for 1KB (0% of capacity); largest unused: [0x2600000-1023KB]
Zone QuartzCore_0x884400: Overall size: 232KB; 7039 nodes malloced for 132KB (56% of capacity); largest unused: [0x3778ca0-8KB]
Zone DefaultPurgeableMallocZone_0x27f2000: Overall size: 4KB; 0 nodes malloced for 0KB (0% of capacity); largest unused: [0x3723000-4KB]
All zones: 36660 nodes malloced - 68691KB

And for 27 images:

Process 1212: 4 zones
Zone DefaultMallocZone_0x273000: Overall size: 167435KB; 30301 nodes malloced for 68681KB (41% of capacity); largest unused: [0x6ea51000-32372KB]
Zone DispatchContinuations_0x292000: Overall size: 4096KB; 1 nodes malloced for 1KB (0% of capacity); largest unused: [0x开发者_Python百科500000-1023KB]
Zone QuartzCore_0x106b000: Overall size: 192KB; 5331 nodes malloced for 101KB (52% of capacity); largest unused: [0x37f2f98-8KB]
Zone DefaultPurgeableMallocZone_0x30f8000: Overall size: 4KB; 0 nodes malloced for 0KB (0% of capacity); largest unused: [0x368f000-4KB]
All zones: 35633 nodes malloced - 68782KB

So what are other reasons for "Cannot allocate memory" and how can I diagnose them? Or probably I made a mistake ruling out mentioned reasons, then how can I check them again?

Turned out I've made a mistake checking that address space is not exhausted. Instead of using heap command I should have used vmmap. vmmap revealed that most of the memory is used by images mapped into the memory.