I always shake my head when I see people looking for the "leanest" software implementation in an attempt to get faster performance:

Another point is speed. . . . A monk can infer that CGI.pm's bloated, slowing down hit requests. Cutting down the amount of fat, running lean, really works.

Really - with the performance of modern hardware, coupled with (lets say you are writing CGI applications) the inherent performance of Apache with mod_perl (you do use mod_perl, don't you?) performance is really not an issue. Of course, if you are trying to run a serious web site on a clapped out 486, what do you expect?

You want to improve CGI.pm's performance by spending a lot of time rewriting a "lean and mean" replacement? Save the money that you would burn in programmer time, and upgrade the hardware.