Capers Jones data on pair programming should make it go away. To summarize, pair programming is much less effective than formal inspections (formal code review), and adds over 50% more in cost.