There are many ways to do this - what I usually do for my testing is using WAIK (the Microsoft Windows Automated Installation Kit, available for free from Microsoft), to deploy a minimal Windows 7 system on a separate workstation.
Then you need to consider / change the following configuration items (do not try not to deviate too much from a typical user machine, otherwise your test will not be constructive):
- Set swap file to static 2x RAM
- Turn off automatic updates
- Disable Disk Indexing
They provide a reasonably optimal testing environment that is still reachable by enthusiasts and thus can be a Power-User representative (even if I use automatic updates and disk indexing, I plan to both of them when I leave / bedroom)
Regarding the use of caches and memory - at least in Win7 Professional, you can script remotely run - so, for example, I would have a script run my test in one night (for large regression tests), restarting the OS after each run. Or I will run the same test 5-10 times without rebooting to see if cache usage is changing.
Finally, there are bootloader switches to control the number of processors and the amount of RAM available. My test computer is an AMD Phenom X6 with 16 GB of RAM, but we need to check how the performance changes with the number of cores (some users will have single-core systems, and some of them will have multi-core systems) and with RAM (from 1 to 16 GB).
This is usually done before the release of the milestone to find out if the recommended or minimum recommendations need to be adjusted due to additional features and additional optimization that has occurred since then.
qdot
source share