There are a lot of "positive network externalities" in the computer world: the more people use something, the more valuable it is. That's true for databases as well.

However, what he's saying is that beyond a certain point, those effects tend to become a bit less important. Java is more popular than Python, for instance, but you can get pretty much anything done that you need to in Python. OTOH, a language like Erlang (sorry to pick on it, it's just an example) is still fringe enough that there are gaps in its libraries, it's tough to find people, and so on.

Echoing your point, having a library that can do something is fantastic. Having two is even better when you have a choice on which one feels more comfortable to use and is more suited for your project. Having 25 libraries that do basically the same thing gives diminishing returns because the community is split over which one is better and easier to use, while the majority won't be full featured and with a larger number of bugs overall.

This is not necessarily a black and white scenario that holds true to everything, however. There is no such thing as an absolute generic.