Do the MiNTLib malloc() call the kernel malloc every time ? I'm absolutely not sure about that. That looks like a problem in the malloc implementation in the MiNTLib. And it is quite complicated.
You're right Vincent, quick look into malloc.c tells me it happens only when mintlib's memory pool has no more available memory. Is there anyone who knows why exactly are here two places where allocation logic applies? Kernel's wrapper for M(x)alloc() does very similar thing so I'm confused now why would one implemented another pool inside mintlib. Is it some general POSIX / OS architecture rule? Or optimization for speed?
Another thing is mintlib (and its debug version, too) has very little trace outputs in this area so it can be no help at all...