have you done some profiling whether this is really necessary? I'm not sure if a self-written memory allocator will really be better than the one in the c library.
i'm not concerned about the allocation of only a few bytes, but of larger sample buffers, bigger than a few kb ... this may cause the dsp to stutter ... if we have this memory already allocated, it's already represented by physical memory (if we use memlocks) so we would only have to assign a pointer from our memory block(s)...
cheers... tim