indeed, but I bet 80% of the time a performance problem is a configuration problem: check that first always and come to this low level tools only after confirming you really have to use them.
Products.LongRequestLogger is something we don't use anymore, as you can get the same information using the tools you already have on your stack, like nginx and Varnish.
a lot of common sense is also needed; I have a pending blog post on how we are using HAProxy now and what problems it helped us to solve.
re-reading the question I see this:
which point to an overuse of this single-thread instance with no front end web server, and probably no caching neither.
the most simple test you can do is just adding another thread to you server:
zserver-threads = 2
note that this will increase your memory consumption by almost a 2x factor.