I've noticed that some heuristics for it on my problem which work surprisingly well. I guess it ought to be systematically studied although I cannot find guides or overviews for it.
It is very much problem dependent. The issue with going from Monte Carlo to Simulated Annealing to Very Fast Simulated Annealing is that one increases the number of tuning parameters that the method has and that are all dependent on the specific problem. The only thing you know for sure is that your temperature schedule must allow for step lengths whose sum is infinite.