The wsgi benchmark uses gunicorn, which is fine. But if you are surprised by the so-so performance, know that the worker class can be changed via the -k flag (-k gevent), and may improve performance. Of course, other wsgi servers, like uwsgi, are also available.
Are they not even using gevent? Someone please get these guys off the internet. The are worse than some personal blogs at everything except the comprehensiveness in their low-quality testing.
