Just stumbled upon this StanfordU paper about ongoing research on the When and Why of Pair Programming (one of the practices of XP):
PairProgramming-WhenWhy.pdf (application/pdf Object)
My personal, not scientifically backed opinion on the matter is that Pair Programming is probably one of the more dispensable practices of XP. Reason: Meanwhile (since the bloom of XP in 1999), there are a couple of other, probably more efficient and certainly more flexible ways of quick and easy programmer2programmer communication like instant messaging, Skype, SIP/VoIP, video messaging, IRC/chatting, collaborative editors. These are potentially less disruptive/disturbing, enabling more concentrated functioning while ensuring an equal or even better communication (due to the higher specificity in peer selection and the availability of both synchronous and asynchronous communication). Besides, the number one communication media should be reading and writing the source code itself (eventually through an editor with on-demand collaborative editing support).