I had fairly bad experiences with pair programming. None of the physical reasons the author brought up applied: it was in a decent, private room, with decent hardware. The social problems did not apply: it was with cool, friendly people.

It turns out that, in pair programming, while the active person hacks with all his neurons firing, the passive one sits bored out of his mind. I found that passively following someone else's programming, and possibly even fixing minor mistakes, barely uses up a fraction of the mental energy of actively coding. The passive ends up being a distraction. When the passive makes a comment ("have you considered changing this function to do XYZ?"), the active looks up, completely distracted and torn from his concentration, and stares at the passive. "What?.. What did you say?.. Oh... yes, that function... I already thought of that, and here's why it can't work." Basically, I found that being passive in a programming pair also blocks the kind of forward-thinking analysis which leads to good programs.

One more thing. The author claims that pair programming helps avoid duct-taping problems. I completely disagree, at least when it comes to my own work. When I look at my code, and I see the beginnings of something bad, I ask myself: "Could someone else understand this in two months? Will I, when it breaks?" Then I go back and fix it. Whereas in pair programming, two people will likely look at each other and say, "Oh yeah, that's an obvious workaround! Let's keep going!"