Here's an idea. Calculate expected performance for all the players who have played for a coach. Do this by looking at their careers before and after. Then look at how many under performed and over performed with the coach in question. Basically do the Mike Smith thing, but with all the players. Now, obviously for any given player there is a lot of chance involved, but with enough players it should regress towards the mean.
The problem with this is of course that a massive amount of calculations have to be made, but I could see it done in a professional setting. For instance an NHL team doing due diligence before hiring.
There's also a problem in how you count expected performance of course, but it seems to me that could be worked out.