[klaruz]

I agree, something seems wrong with that test. Varnish performance should not top out with the exact same performance no matter the concurrency. Here's an older article that shows with some very basic changes you can get 27k/s on an old single core machine: http://kristianlyng.wordpress.com/2009/10/19/high-end-varnis...

[todsul]

It's a totally different architecture. It's adding another layer. Why isn't it conceivable that adding this overhead would reduce speed?

Remember, I'm not comparing a Varnish cache to an un-cached site. I'm comparing cache vs cache. Nginx serves static files very quickly. And cached files, in this case, are static.

Also, these are single-server tests. Everything changes when you start talking about very large websites.