I wonder if anyone ever used Erlang distributions for website performance metrics.

His math should apply and offers an effective, statistically valid method of estimating resource requirements, much superior to the usual "requests per second".

Actually this is explicitly applicable to the web/Internet performance. The person who discovered this distribution was the same person for whom the language Erlang is named after-- and was developed for the purpose of predicting waiting times between telephone calls. This also describes times of web page requests arriving to a server.

Much of what flowed out (queuing theory) can just as well be applied to Internet packets or to web servers (with various threading/request handling models corresponding to various queues) processing requests.

This may not sound exciting, but it's vitally important (particularly for modeling and simulation-- which is vital in cases where you can't do a live test of a certain algorithm without incurring an expense).

Fundamentally these algorithms are fairly easy to implement: all you need is a prng producing the specific distribution. I wonder how many of the HTTP stress tools actually do this, though.

Wow, this is so boring I flagged it.

This is the distribution you use to analyze people waiting in lines! Unless it doesn't fit right in which case you just use another distribution.