They've got a ton of really great tests. Seeing the actual change in revenue is a crazy - my favorite test was 4.2, where they added some ads to the homepage and decreased revenue.

Overall an excellent paper that both discusses specific, relevant examples as well as critical aspects of experiment design.