Mike Cohn cited a statistic that pairing added 15% schedule overhead above solo programming for the construction portion of a feature. Not sure where it came from, but it’s a pretty compelling argument for pair programming, considering the intrinsic benefits of knowledge sharing and improved quality.