I use multiple worker processes in nginx (2 x # of cores), which work in parallel. Then FCGI::Engine is configured to run multiple processes as well. Finally, I use the external JS and image files (in part) to speed up the client-side via parallel processing of requests. (Note that IE7 and below limit concurrent connections per host by default often to 2: http://msdn.microsoft.com/en-us/library/cc304129(VS.85).aspx).
