Thanks! I think that answers my question. Since kernel threads will likely have 64KB stacks or even more, this is still a huge win, even for 32-bit code I guess.