>How do you do repeated SQL performance tests and not be deceived by the caching issue?
with data that small only a reboot will truly be enough if working with internal laptop disk.

Most of the type of work I am doing it is more relevant to measure the performance of
already cached data, so I usually measure only after a few dry runs to fill the cache.

Is your query hurting your start-up time ? Why is the "cold" time more important than cached perf ?
I'ld go for the "typical" scenario, and if the query is often run, data should be in the OS cache at least.

One thing I haven't tried: if you are working with multiple disks and removable media
(USB disk or Hot-pluggable SATA fall into that category!) remove the media not used by OS and re-connect.

Keep me posted of your findings

thomas