If Netflix doesn't keep social graph data, then they won't know the probability of you ending up in a conversation about a particular movie. That alone might make a lot of people choose the social recommendation.
But keep in mind, to be valuable the social recommendation doesn't have to be more accurate than the anti-social recommendation. It doesn't even need to be chosen by more people. All you need is for some people to prefer it and it's valuable.
But they can take the many movies you rate, find people who rate similarly (which may be way closer to your taste than your friends') because of the large data set.
Making money off of it and justifying a $15 billion valuation does require a lot more than for some people to prefer it.
Also, even if they did beat Netflix at recommendations, how does that monetize?
I won't try to justify the $15B valuation. That's obviously speculative. I just agree with Paul Buchheit that there's interesting potential in the social graph.
Back to netflix, I do think the anti-social graph will work better for some people, especially independent thinkers whose tastes haven't changed lately. For others, what movie they want to watch may depend on what movies their friends want to watch. Or they might have a change in taste correlated to a change in social circle. In those cases the social graph would work better.
To monetize, friendfeed could charge commissions for "get it now" links. They could sell ad space that is as lucrative as search ads.
But keep in mind, to be valuable the social recommendation doesn't have to be more accurate than the anti-social recommendation. It doesn't even need to be chosen by more people. All you need is for some people to prefer it and it's valuable.