I don't know about any broader statistics, but in my personal experience, I see all three of those. I think it's mostly a function of how large the user base is, how complicated the code base is, and whether or not there are any substitutes.