Hm, you may be right. The methodology to arrive at these results in the paper is not given at all (as far as I can tell). However I suspect that the three you

Hm, you may be right. The methodology to arrive at these results in the paper is not given at all (as far as I can tell). However I suspect that the three you listed would not be benchmarked on their own, as they are not application servers, just frontends. It would be more reasonable to benchmark Apache+mod_wsgi or whatever, and it would be nice to see that on the graph.