While mod_perl and the like are great and all I would really strees looking at the speed the code it self exicutes.
In most cases the server process will not send any data, not even the HTTP responce headders, until all data from the CGI process has been read.
That means a pourly writen loop, even after most of the page has been sent to the server process, can chew up valuable seconds, especaly if you have any SQL querys inside a loop.