That's a very misleading graph. It's not about how many people read tweets through the different channels.

There are many reasons to log into twitter.com, and so the graph needs to break out (or at least explicitely state) how many people read tweets on twitter.com.

Another, more important, point: there is probably a long tail effect here. I have a hunch that when you aggregate all reading via API vs reading via the web, the breakdown would show a huge API to web ratio.