The best advice has already been given: test the scenarios rather than guess. Even the most experienced coder can fall victim to bad guesses when it comes to optimization.

However, since you asked, I'll offer a bit of a guess anyway. First establish that you have a performance problem. If you don't yet know if you even have a problem, you are prematurely optimizing; and premature optimization is the root of all evil. If you don't truly have a problem, let your design decisions be guided by some other important factors like maintenance of the code, future growth, re-use, etc.

My guess is that under mod_perl, the performance difference under all of your scenarios is likely to be statistically insignificant. Other issues like network activity, SSL, possible database access, etc. will have a much bigger performance impact than similar code divided up different ways.