Using this limit on GitLab.com it appears we're able to reduce response
timings by about 620 milliseconds compared to the previous limit.
See gitlab-org/gitlab-ce!2421 for more information.
Don't assume that if the Rack server is not Passenger then it must be Unicorn. There are many other Rack servers in the world (uwsgi being one example that people use a lot).
The reverse check is much more logical, i.e. check explicitly for Unicorn
