>Finally, I recently upgraded my main work system from a PII 333 to
>a PIII 600. It seems zippy but not as fast as I expected: My last upgrade,
>from a PI200 to the PII333 seemed to make quite a bit more difference.
>The new system runs a simple timing program a little less than twice as fast as
>the old, but doesn't seem twice as fast overall.

Yes, there is more to system performance than CPU speed.

The following is intended as useful background info.  The specific solution
to your problem involves a set of cost/benefit decisions which only
you can make.

In a nutshell, you cannot have too much cpu,ram or disk and all of them
suffer if you don't have enough bandwidth between them.

In general, a fast CPU is good (faster is better :-), but a slower CPU with
more RAM will often do a better job.  And since no amount of RAM is ever
enough you will eventually start paging (or worse - swapping), at which
point your disk I/O starts to matter.   

When configuring a system for performance, we want lots of disk spindles
and lots of controllers to spread them accross.  That gets _very_ expensive
very quickly.  All these things are good though - the trick is dividing
your budget up to get a fast enough CPU, enough RAM and disk spindles.

For single disk configurations, modern IDE disks are supposed to be 
as good as scsi (I don't use IDE so don't quote me :-), and $/Mb is much
better for IDE than scsi so its hard to suggest anyone go scsi for a 
single disk system.

However, once you start wanting multiple disks scsi is hard to beat.  
You will of course be limited by the number of PCI slots and bus bandwidth, 
but since they aren't adjustible we can ignore them - though the 
beauty of NetBSD is that you can change your hardware architecture 
without changing your OS :-)

--sjg