Alex, your problem sounds very strange. The repository URL is stored in your account and passed to the specific room when you start a recording so I don't see how just setting the URL again would fix the problem.
A couple of things I'll need to check (but this week we are all off) is what exactly happen when a room with a recording in progress fails. The room will restart on a different FMS but I start suspecting we may not restart the recording again (one more reason to have the recording flag as a node item in the room).
For sure currently we are not doing anything to recover the initial part of the recording. This is something we debated a long time: in a previous incarnation of FMS recording we used to push the recording data to two different servers in real time, so that you could recover the initial part if one of the server would fail, but on top of being an ugly implementation it was also hard to put back together different parts of a recording, so in this implementation we decided to keep the files local to only one server and push it to you when the recording succesfully ends. It may not be the perfect solution but the previous one caused more problems than it solved.
Anyway, to get back to your problem, do you have any trace that would show if your room got "bounced" between different servers ? Or if you can send me the room URL and an approximate time of your recording I will check our logs and see if I can at least figure out if the room showed any problem (but again, it may be next week).
Thanks for your patience! (and for helping out in the forums)