This is a good point. I actually believe the big time hog is the large number of separate (and in some cases rather complex) queries made on the backend database. There may be some issues with passing/assigning arrays/hashes, but I've quashed alot of things like that already with no humanly noticeable improvement in speed.

As for the impact of prototyped vs unprototyped sort routines I timed a portion of the code that does a lot of sorting with each version. The prototyped version added about a hundredth of a second to execution time (1.371 vs 1.382 seconds). So I've gone with the more straightforward prototyped implementation, ignoring the minute slowdown.