Several times the topic of rewriting the ceval loop
(eval_frame) has been brought up as a possibility of
speeding up python.  Here's the most recent.

http://mail.python.org/pipermail/python-dev/2003-February/033680.html

Quick results:  after the rewrite, Python was about 15%
slower.  I think a bunch more optimizations could be
made, but I think the best case would be to rival the
existing speed--so I stopped.

The way to squeeze more speed out of this patch would
be change the calling conventions.  One note, this
patch probably uses less memory because there were a
bunch of local variables removed from eval_frame.  Of
course, some are also included in the ceval_struct.

I removed the entire switch from eval_frame.  There are
a bunch of snapshots along the way included in the tar
file. 

All this work was against Python/ceval.c revision 2.350.