Plus instead of running 16 Unicorn workers for an app, eating up over 4GB of memory, I can run one JRuby thread-pooled app, handling many more connections gracefully, with lower memory requirements, fewer database connections and better performance.
Very pleasant surprise indeed.
You get the whole Java ecosystem at your fingers. It's faster than MRI (1.8.7ish) and a contender (maybe even faster now) against MRI 1.9.2. It interops with other JVM languages. It plays nicely with uptight IT departments (you only need to download the jars).
Maybe it was behind the scenes, but the original status pages for YARV gave the impression it was a one man job up until it was official.
(Although i can't google any examples in print)