A fresh out of the oven Squirrelfish called SquirrelFish Extreme was announced here. They attribute their speedup to bytecode optimizations, Context Threading, a property cache, and a regular expression JIT.
I tried a context threaded interpreter in Tamarin-Tracing, which significantly sped up the interpreter, but jumping from a trace back into the interpreter was too expensive. David Mandelin has been looking at adding inline threading/context threading into SpiderMonkey. Something like a property cache is in SpiderMonkey and has given SpiderMonkey significant performance improvements. Finally, we were thinking about doing a regular expression tracing JIT, but have yet to implement it.
Cameron Zwarich already has some performance benchmarks of SFX (SquirrelFish Extreme) to TraceMonkey and Google's V8. Browser war round 2! FIGHT!