I can't confirm this with the stock Perl 5.8.6 on OS X 10.4 -- the second run seems to be consistently as fast as the first.
However, I have experienced similar phenomena on this system before which I figured were a result of memory pool fragmentation. I have an indexing benchmarker script which inverts around 20000 Reuters news stories. In the initial version, once the script had been run a time or two (warming up the OS hard disk read cache) the first run was consistently faster than all subsequent runs by a few percentage points. I wound up launching a child process for each run to defeat the effect.