call based vs continuation based contification

[Stephen Weeks]

> By  the  way, I see that for count-graphs, the ultimate benchmark, using both
> call-based and continuation-based was a 26%  speedup  over  just  call-based.
> Definitely not too bad.

Yes.  In count-graphs, there were two mutually tail recursive functions that
should have been contified but weren't because one called the other twice.

> You never said what you thought the problem was with the cases where both ran
> slower.  Any guesses?

I looked into knuth-bendix, but couldn't find anything.