And your controlled experiment is what exactly?
A secret. actually, it was a plan for a blog entry. I already have the VM sort of set up, just have to be bothered to actually perform the experiments with my XP snapshot.
The idea is this: I create a setup with a Clean XP Install, and test several "registry cleaners" on it; after each one, I revert to the snapshot I took of the clean install, and try another.
I also have a "loaded" system, where I uninstalled, reinstalled, deleted without uninstalling, added back new versions, deleted old versions, and basically tried to muck around the Install as much as I could to "simulate" what might happen over a longer period. Again; each time I will test the cleaner/fixer tool, and then revert to the snapshop I made of this "loaded" XP install.
For each one, I use Process Monitor to determine exactly what the cleaner is doing; most illegitimate ones (one on my list being RegCure) will likely show that they hardly access any registry keys at all and simply report specific keys as invalid using a time delay.
Additionally, I'll ensure that the programs installed work properly- in all respects- after running the cleaner.