There's a much higher bar, too, to any form of research that involves human subjects. There's a whole regulatory regime that must be complied with to even get permission to do this kind of research. If the possible downside is high (as it is here... what if the treatment makes the subjects
worse and a bunch of them wind up more suicidal than they were before?) the level of required oversight is pretty high.
I'm not saying that's a bad thing -- unethical and/or misguided researchers have actively harmed their subjects too many times in the past. Nonetheless, it's something that people who haven't done human subjects research often overlook, and something that can add a huge amount of expense and delay.
If you're doing generic CS research* the worst that usually happens is that you have to reboot. We can't reboot people, at least not yet. :-)
*If you're doing research on, say, human computer interaction, you generally do have to get human subjects approval, but not usually for compilers, algorithms, or the like.
Edit: "people willing to take an untested treatment"
That's exactly the rub. The potential subjects for something like this are what's known in the jargon as a "vulnerable group". You have to be very, very sure (and be able to convince a skeptical oversight committee) that your subjects are, in fact, giving their active and informed consent for an experimental treatment. That's quite hard to do for people with severe psychological problems (again, rightly so... some of the abuses of the past make for sobering reading).