I will start doing some testing once I will get the test server and see how many SPs are touched by users, maybe it would be indeed a small number and that will get me close to the benchmarking that MS done for 2000.
The benchmark should be fairly close to my situation since the memory allocation for the procedure cache is similar in 2000 and 2005, as you guys pointed out.
Thanks for your help guys, I will keep you posted in case any of you might encounter the same situation.