Thomas has some very good questions. So far, this has not come up because we have not copied GNU FDL materials into the database. (Oh, except I did see some entries from a dictionary of computing that the owner gave us permission to use.)
It's a difficult question! We selected the GNU FDL license because it is indisputably "open source" or "free" (choose your religion! ;-)). But to a certain extent, it is "tighter" than we want it to be, as per the provisions Thomas mentions below.
Notice that all of the provisions mentioned can be omitted if the original author doesn't include them or insist on them. Within Wikipedia, we have generally had a strong social norm against the concept of "authorship". What I mean by this is that it is considered bad form to get annoyed at someone for changing "my" article. The best entries are semi-anonymous -- I mean, your name is there in the change log so people can thank you for your excellent work, but the article itself stands on its own.
Other little puzzlements of this kind are in the license if you mull it over. What is really needed, I suppose, is a license that is compatible (in some sense) with the GNU FDL, but which doesn't impose all the restrictions of the FDL.
The one thing that we don't want is to say that the stuff is in the _public domain_. The reason we don't want that is that it makes it possible for someone to take our work and make a _proprietary_ version simply by modifying it. I think that by and large the contributors wouldn't want that -- and I know that I wouldn't.
Also, we are currently using the "invariant section" bits in the license to require people like Yahoo or Altavista who may mirror our content to provide a link back to the main project. (This has not actually happened yet, but we hope that it will someday.)