There is always the dilemma between allowing users to setup their disks as they want, one for project, one for media, one for exports and one for previews, or making it simple by using only one disk for everything.
The dilemma is the script questions to ask about all the other locations and their characteristics. For most people the current 22 questions is already a burden, or at least a bore. Increasing the number of questions, that only applies to less than 50% of the participants if we look at the PPBM4 results, did not seem a good option and in our effort to KISS, we decided to leave out that option.
If you have an investigative mind, you can manually modify the script to reflect your setup, but I would advise against it. During all the testing we have done, and Bill and I have each done multiple times the tests that are currently submitted, we have not found any sizable difference in results. This is due to the fact that the test methods have changed drastically in comparison to previous versions.
Rendering is only done once per test, so the location of the previews is rather irrelevant and the previews generated are not used in all the other tests. For the remaining tests, Disk, MPEG and H.264, it is easy to export to another disk and modify the script accordingly, but apart from very esoteric disk setups like Studio North (9 x R3 and 5 x R3), the benefit will be marginal, especially on the two CPU intensive tests.
If you look at the Disk test, which reads a lot of small files and writes one large file at the end, the read disk has long finished when the CPU has finally collected all material for writing, which is done in a single burst. It does not matter if that is written to a single disk A or a single disk B, because their burst rates are pretty much the same and any differences will be below the measurement accuracy anyway.
Hope this explains it a bit why we chose for the KISS approach.
If you want to test out different locations for files then go back to PPBM4 and you have the options, it works perfectly well on CS5