It is non-blocking and executed within Nginx itself so it is supposed to be very fast. The Lua work has been done by Chinese engineers from Taobao and Weibo which obviously get a huge amount of traffic.
It is a very fast module, and non-blocking, but just as importantly it provides the flexibility to avoid sending requests from Nginx to slow backend servers (Ruby, Python, whatever) at all. In real-life situations, this is the most important bit.
As I said in another thread, we've done some good work to tie this in with our Rails app. We now page cache almost every important request, even for logged in users. This means Rails and our databases are protected from most spikes of traffic. I would like to write a post describing this in more detail, I need to think how to get it across though.