I don't want to imply that there's a hard-and-fast rule here -- there isn't. And it's possible (but highly unlikely) that I would hire someone who is entirely self-taught. It's more of a spectrum: if you've done very well in a program than I'm familiar with, I have a great deal of certainty about the kinds of problems you've dealt with; if your degree is slightly different (EE, CompEng) or from an institution that I have no familiarity with, that just means I need to familiarize myself and wade into the specifics -- it's not a strike against, by any means. If your degree is further afield but technical, that will require more validation that you understand computer science; I've seen way, way too many physicists who turn out to be horrifically bad software engineers. If your degree is non-technical, however, then you're going to get lumped into the "self-educated" camp -- and again, it's not impossible that I would hire someone self-educated, it's just highly unlikely to encounter someone self-educated who meets our bar for software engineer.

One of the things that I've noticed in this discussion is that there's been little mention of what your company does (I do, I believe I'm citing some of your work in my M.Sc. thesis). While, in general, I suspect that self-taught developers can do a lot of great work, I also suspect that the kind of development that you're doing at Joyent would very strongly benefit from formal CS training.

I understand bcantrill's point as a quest to remove uncertainty. He wants to interview people who meet certain basic criteria so he can focus on what makes one CS grad different from the other. If he interviews people with more diverse backgrounds he will have to deal with more diversity and the interview would consume more time and resources.

But an engineer or a physicist would probably be able to advise him on what he should sing his hard drives to make them run smoother. I don't think a compsi grad could do that ;-)