The misleading graph doesn't help their case, either - tests were run at 1,5,10,25,50,75, and 100. The graph mistakenly spaces them evenly apart. In reality, twisted.web died after 50, while Tornado still worked @ 100.
Also if you're planning to do any real Comet stuff, this is an extremely small-scale test. Should try with a few thousand concurrent requests to gain any real insight.
