I'm currently volunteering as the head of the computing dept. of a student radio station, and this year we've gone completely the opposite way.
Why? Because when I arrived, we had a server called "*name of station*fs1" (File Server 1) which wasn't a file server, a server simply named "*name of station*" (makes for fun times when it goes down...!) which wasn't the main, all-powerful server, "jukebox" (which did run the station jukebox... and more) and some other systematic, role-based names such as *name*sw0 and *name*backup1 that'll probably stick better but who knows. Such names just don't work well when the workload moves across servers, we've found, which is often.
Then we got a new server for running builds/development stuff, and I decided, in an optimistic prediction of its stability and uptime, that it was going to be called bsod (Building Server of Development!). This sort of name still works quite well in non-backronymed form now that bsod is the main, generic Computing Team server. Since then, we've decided as a team that naming things with role-based names == setting up an artifact name bomb, and recent naming conventions have included naming servers in honour of well-loved Computer Science department lecturers, "dog" (of HELLO, THIS IS fame) and at one point a pair of servers was going to be called "red" and "blue", after a famous pair of handheld console RPGs...
Having said all this, the fact that we're run primarily by undergrad students and have a small number of servers means that we can get away with giving things silly names. We'll probably start CNAMEing/HOSTSing the role names eventually (already do this on the HOSTS level for the database server), so it doesn't matter if "dbserver" is "*name of station*specialpikachuedition" or whatever, and we can repoint the DNS if things move around!~