To start off, here is the distribution of frames per second (FPS).
A surprising amount of people have less than 30 fps. (The 30, 60 and 120 fps spikes are due to VSync.)
To narrow down what may have caused this, and to see if it could be prevented, I decided to graph the correlation between the fps and various settings.
Since all correlations are below 0.3, there doesn't seem to be any one setting that affects the fps in any major way. This means there's no simple magic button you could toggle to increase the fps, but that's probably a good thing since it means there's no major performance hogs.
Even looking at settings which you can't easily change, such as the number of CPU cores, there's no setting which goes higher than 0.3, which is well within the window of random noise.
 When I analzed the average fps for a varying amount of minutes played (defined as the number of logged records per unique snooper token * 15), it turns out none was below 40 fps, which means the users with less than 30 fps was probably just having lag-spikes.
How many people are playing in windowed vs fullscreen mode, and how does it affect the time played?
This is kind of surprising. People play about 10x more in fullscreen mode, but windowed mode is far more popular. This may be selection bais, since people who know how to turn on fullscreen mode may have also been playing the game longer, and are thus more likely to play for longer.
How many people are playing in the applet, and how does it affect the time played?
This is even more surprising. People who use the applet also play for about twice as long, despite desktop more being more popular. I'm not really sure why this is, and it seems hard to belive it's an anomaly since two million data points was analyzed.
Disclamer: @minecraftcpw Notes that this statistic may be somewhat incorrect.
There were a lot more vendors than these 4, but they were all used by far less than 1% of users, so I didn't include them.
Click here to read the follow up analysis to this post.