I’m firmly in the “database over tagging library” crowd, but I honestly don’t see the big divide.
A good, well populated database can have its data be manipulated to make for exactly the tags someone using it for tagging wants, but if you want to make a linguistic paper on how often artists are credited with “ft.” vs. “feat.” vs. “featuring/featured”, you can’t use our current data as you can’t go the other way. And some people who tag want to use exactly what was written on the cover anyway.
One of the things I think is really important about MusicBrainz is that it stores the data and makes it available. Some people will then use it for tagging. Some people will use it for academic research. Some people will dump the SQL and print it on a paper and feed it to their cat. Some people will make a service matching MusicBrainz data with OpenStreetMap (or similar) to provide guided, interactive tours of music history of any given area. Some people will make an application that lets you determine the “degrees of separation” between two music people.
As long as you have enough quality and quantity of data, there are so, so many things you can do with it. Tagging your music/audio files is definitely one of them.