I'd like to see the difference in bug detection between "pull requests" vs. "over-the-shoulder code reviews".

Edit: or, rather, in my experience the author of the code often finds the bugs when explaining it to the person standing next to them. This goes away in the GitHub-style pull request.