One thing that dynamic optimizations have a problem dealing with is eliminating the overhead of indirection in data representation, e.g. from arrays of objects

One thing that dynamic optimizations have a problem dealing with is eliminating the overhead of indirection in data representation, e.g. from arrays of objects or nested objects. If all of the objects involved have a limited local lifetime, then you can just inline all of the functions involved and apply standard optimization techniques. But if the objects have a more global scope, then a JIT really can't do that much. There is some research into this (see http://ssw.jku.at/Research/Papers/Wimmer08PhD/ http://ssw.jku.at/Research/Papers/Wimmer08PhD/ for an example), but as far as I know nothing is production quality.