>Thomas & Al,
>
>You're both right. The dual Xeons are hyperthreaded, making them appear as 4 processors.
>
>To make this even stranger, I tried running it on a less beefy server (Single 3.0Ghz P4 (non-Xeon, but still with HT) with 1GB RAM). It beat the dual-Xeon machine by 20%. Isn't zat vierd?

vfp usually reflects cpu speed quite well and just incrementing a local counter var should be running inside the cpu cache. If you are incrementing a property marshalling across COM boundries on each call all bets are off <g>... I worked on double CPU servers, double HT-CPU servers (but not very intensive) and double core CPU's, and that margin smells like "other factors".

There are still a few tweaks in the registry, but I don't not think they result in such a discrepancy.

my 0.02 EUR

thomas