OK, so the problem on OS X turns out NOT to be related to mmap size limits, but simply a problem in creating and opening the swapfile. The swapfile does not get created properly on OS X and thus mmap fails later on, as there's no file to mmap to.

OK, so it turns out my "fix" just produces another crash; the fix will be then to really fix it somehow:

@alvinhochun Yes, good point, I'm currently headed for holidays, but I'll stop by #krita when I'm back.

Actually, my calculation above was wrong; it's 0.3% for one thread, but since that code ran in 7 threads in the profiler (all summing up to 100%), the impact is more like 0.3% * 7 = 2,1%.

@alvinhochun is right, I missed that.