(Looks at comments) Sigh.
Okay, everybody who has actually mob programmed? Raise your hands. Let's hear how it went. Everybody who has an opinion -- usually about how it's all a crock of shit and a waste of time -- you guys to the back of the room. No offense, but this is kinda the standard response to anything posted on HN.
Modern programming, especially at BigCorp, is much more about tooling and environment (both technical and social) than it is technology. People who get together and work a piece of value from front to back in the system quickly gain a common understanding of how all that tooling and environmental structure is supposed to work. In a day or two of mobbing, you can take a team through what might otherwise be months of mentoring and code reviews. It's so effective at teaching "how to make solutions here" that I'm hearing stories of noobs picking up an entire language simply by participating in a mobbing group over a few weeks.
Got a team that wants to make mob programming into another committee meeting? Guess what? Your team is broken. Get that problem out in the open on day one and then fix it. Don't hide behind process. Mobbing makes you do this.
If programming were measured by the amount of keys you pressed, or KLOC, mob programming would make no sense at all. But it's not. Programming is about making computers do things that people want, whether it's one line of code or a million. So what you're optimizing for is the appropriate mix of diverse problem-solving methodologies and personality types in order to maximize the quality of creative responses to the business problem.
I'm a big mobbing fan -- having watched it and done it. I don't think it's a magic bullet or appropriate everywhere. I think the interesting part of this discussion will be identifying situations where it's a bad idea. (And of course that would be done empirically, not anecdotally)