Ahh, I see from doing a bit more research that JIT does *not* run  
particularly faster than statically compiled code. But rather, it runs  
faster than interpreted byte-code.

I remember many years ago speaking with David Robson, over lunch,  
about the upcoming changes in Smalltalk, using a form of JIT to  
improve performance of their method dispatch, and attempting to gain  
multiple inheritance in that manner for Smalltalk. But there, again,  
it is a case of attempting to improve on an interpreted byte-code, and  
not a case of improving over statically compiled native code.

But with so many talented bodies working on LLVM, perhaps, in time, a  
way will be found to gain improvement over static native code.

Dr. David McClain
dbm@refined-audiometrics.com