> OTOH my 'gut' feel is that the pool allocater isn't as
> efficient as malloc.  Especially if malloc:
> ...
> - had an interface for fixed sized object where the allocation
>   list was pre-determined

This is actually very easy to do transparently, using GCC's
__builtin_constant_p() and some macro hackage.  (C.f. what I did in
the i386 pio.h and byte_swap.h.)

However, traditionally this is what the MALLOC() macro was intended
for.  It's not really used any more because we keep adding options
that defeat it and enabling them by default...