I'll try to reply to few previous post here.
I didn't test swiftyply very much, and i didn't do benchmarks comparing to nginx so i have no idea how speed comparsion looks.
But i DID test nginx vs. apache vs. apache2 vs. light and nginx always wins 
now mongrel vs. evented mongrel - evented mongrel wins BIG
clasic mongrel - forking mongrel (my own modification that makes mongrel act like apache - fork on request) - a little bit slower then classic, but for low traffic sites - much better choice then cluster - you only need one mongrel instance in your memory
mongrel vs. nginx on static files - nginx wins - but it's power is only needed when working with lots of small files. i also have moded mongrel that is designed to handle static pages - it allows caching most used files in memory - like css styles and main page, i'm planing to improve it with sendfile syscall for speed on files that are not in cache and mmap on things that are in cache.
BUT all this improving is only for my fun and pleasure becouse you are more likely to run out of bandwidth before speed will be an issue
now frameworks - i didn't test all and mostly I'm basing this list on my old benchmarks, what khaines said, and other things found on network
Order from slowest to fastest
- rails
- nitro
- ramazee
- merb - almost 15 times faster then rails in my tests
- iowa
somewhere between merb and iowa probably would be my micro-framework, still unreleased - but i'm geting into it and it'll be availble this week i think for alfa
now ORM
- AR
- Og
- Sequel / Kansas - no idea which one is faster i didn't test them
sequel is 2-3 times faster then AR on small and frequent queries, but have nice features like connection pooling for postgres which gives it even more advantege.
Okey that would be all in terms of speed, but remember that usually lower level you work on - more time you spent and more bugs you make, we were considering using somethning other then rails in our projects and we found out that if project doesn't REALLY need speed, costs of adopting new technology, training, testing etc. exceed what we would save on hardware.
cheers