Code Reviews: Just Do It (but really just a quote from Steve McConnell) "...software testing alone has limited effectiveness -- the average defect detection rate is only 25 percent for unit testing, 35 percent for function testing, and 45 percent for integration testing. In contrast, the average effectiveness of design and code inspections are 55 and 60 percent."
This is found in "Chapter 21: Collaborative Construction" of the 2nd edition of "Code Complete". Furthermore, he goes on to say that "The cost of full-up pair programming is probably higher than the cost of solo development - on the order of 10-25 percent higher - but the reduction in development time appears to be on the order of 45 percent..." The defect detection rate, stated in the book, is listed as 40-60%.
It also includes some tips for successful pair programming: coding standards, active, frequently rotated pairs, a good pairing environment, etc. Of course, he seems to think that pair programming involves one keyboard, one mouse and one monitor - which doesn't seem like a very good environment at all.
The hard data section contains lots of references to the benefits of early detection of defects, which are also mentioned in a paper, "Planning to Get the Most Out of Inspection". Similarly, "Testing is one of the Least Effective Defect Removal Techniques".
From a recent XP meeting.