The failure mode I usually see on experienced pairing teams is overly complex code. There's a certain flavor that I can usually pick out and be pretty sure that a couple of Pivots or Carbon5 or whatever pair shop has been working in a codebase. I wish I could be more specific about what sets me off, but I've called it blind enough times to think that there's a real effect.

Not that single coders don't produce weird code, just that I've not experienced a situation where pairing has been a win—with the BIG exception of doing code reviews.