@headius knows what this is all about.
Is there any way I can reproduce this? There are now builds of the new invokedynamic logic in OpenJDK, and we could help the Hotspot team by testing your stuff out and reporting performance numbers.
I think I cannot reproduce it anymore on latest JRE (u5) and JRuby. I'll reopen the bug if it will remanifest itself. I'm regularly running performance tests on JRuby head.
Great, thank you. If you're interested, I can get you access to the new invokedynamic logic slated for Java 8 (and hopefully Java 7 update 8), which has not yet been optimized but which should eventually optimize much better than the current logic. Find me on IRC if you're interested.
We'll call this fixed in RC1.