The DLR’s Adaptive Compiler
The Dynamic Language Runtime has significant performance improvements over traditional interpreters for Python and Ruby, once it is warned up. But for code you only use once or twice, it costs more to compile the expression tree than to simply interpret it.
The DLR solves this issue by introducing what they call an Adaptive Compiler. Normally the DLR will interpret expression trees instead of compiling them. If it sees a given tree is used repeatedly, then it will compile it in a background thread to be used by future calls. According to Microsoft, this is giving IronPython 2.6 a significant performance boost over earlier versions.
Because this technology is backed into the DLR, any dynamic language targeting it may opt-in for this behavior.
I'd generally take benchmarks with a pinch of salt, but I have to say I think they have some more work to do on the DLR yet ;)