In the comments, the authors give the sample size as about 20 in each group. Here's a more sensible graph (error bars to create a 95% CI on the estimated group means):

https://skitch.com/johnjosephhorton/fih62/quartz-2

Obviously, there are some statistical power issues. It's unfortunate that they didn't get someone to help them with a power analysis up front---given how many very tech savvy people are interested in Kahn Academy and want to see it succeed, I think they could have gotten this sort of assistance pretty easily.