That doesn't seem that unrealistic — you could allocate your stacks using slab allocation, for example. I wonder why the Kernel allocator doesn't do a better job though.