As to why, here is a hint: The three most common webservers in the world are Apache, nginx and IIS. Yet are any of those even mentioned in that paper? NO.
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.
As to why, here is a hint: The three most common webservers in the world are Apache, nginx and IIS. Yet are any of those even mentioned in that paper? NO.