Vincent,
Yes you're right of course. Even with a cache like mine that eliminates reading of files, and even most memory copying and allocation operations prior to executing scripts, you're still going to be even faster if you simply cut out chunks of code. I do this extensively on a site, and it of course makes a difference. Combined with transparent caching of static query results and on the fly generation and caching of more optimal page rendering code, it makes a big difference :-)
If you're hitting the disc to read your source then of course there's a chance for I/O contention with whatever else is going on - another good reason to use a cache btw.
This is why you should:
1) use good quality and fast RPM drives
2) ensure that UDMA is turned on with hdparm (make sure you have UDMA drives first please ;-). On most linuxes it isn't by default, and people never even realise that their lovely fast UDMA discs are giving them about 3MB/sec performance instead of well over 30MB/sec that they should be getting!
Why not try hdparm -t -T /dev/hda today and see what you get. Anything less than around 25MB and you're I/O performance needs help. (160MB/35.5MB mem/disc performance or even better is attainable)
3) Do the best thing you can do for any machine - give it some more RAM. Running with 128, then go up to 256. Got 256, why not go to 512. Your box can probably handle 1 to 1.5G Ram, and even fast ECC ram is pretty cheap. Time spent on trying to tweak scripts and then on fixing what you broke in the process is not.
Nick