An old coworker of mine at an internship did exactly this to find the most attractive profile picture for a dating site. He had a sample size of maybe 10 photos and used HotOrNot to do a simple ranking of them. The top rated ones were used as his profile picture on social networking sites.

Unfortunately, he left out one part of the experiment - he did not have the "ugly" photos as his profile picture, measure response rates, and then try it again with a "hot" photo.

I am currently running an experiment with my OkCupid profile along similar metrics - I am varying my profile picture, as well as certain parts of my profile, and testing profile views, response rates, and 4-5 star ratings. This is mostly out of frustration with the pitiful view/response rates I was having originally, but if the data is interesting I might do a write-up.