I would also like this to be less than the slowest thing possible, but unfortunately my experience with x86-on-python has shown that to be tricky.