I'm doubtful that a JIT could realistically adapt its output based on runtime performance metrics, but if you have links I'd love to read more.