The patches do not make the box "uber faster for its purpose" no matter how much I'd like that.
Con has concentrated on the end user experience, perceived speed if you like. So he advocates boosting priority on processes which are 'interactive' i.e. the program you are working with. He has been lowering latencies and avoiding swap usage and reading ahead data. He also does not like stalls in his IO-scheduler, therefore he chooses the Fair 'CFQ' scheduler as default (instead of AS). While these measures might improve your desktop experience, I seriously wonder if your TPC's would improve.
He has really put a lot of effort into the Desktop experience and has released a server version of his patchset in which he disables the interactivity bits and rules out some things which do not make any sense on the server like having HZ=1000 for the process-scheduler. However, Con hasn't got big iron in his room and I would not (yet) advocate his server patch for any server.. Just as much as I would not use 2.6.17 vanilla in a production environment.
That being said I like Con's patches, I run it on my desktop and squeeze it even a little more.
