Here are the timings for my machine. With this type of program, the results will depend on the relative performance of the hard disk, the CPU, the memory, and the caches.

Even on a single machine, the results depend on the operating system kernel, and most importantly, with the build of perl. My vendor perl is an unusual build, and my own build is more tuned to my machine.

The kernel I have been running, 2.4.22-6.ll.rh80, is an experimental low-latency kernel used for real-time applications. The file system is ext3.

Timing, seconds	Kernel 2.4.22-6.ll.rh80	Kernel 2.4.18-14
Vendor perl, loop	2.938	2.569
My perl, loop	1.957	1.677
Vendor perl, Storable, no /i	1.267	0.966
My perl, Storable, no /i	0.687	0.567

If you prefer lower case to make the output look better, just use the lc function instead of the uc function.

I think the lesson is that performance improvements are often difficult to generalize, so you have to try it on your own system and see what works.

UPDATE: The Code Smarter suggestion is very good. For the Storable code, timings for the vendor perl on 2.4.22-6.ll.rh80 the timings are down to about 0.21 seconds and with my perl they are at about 0.13 seconds - too fast for my primitive benchmark.