Well, no errors should cause the XData server to die. I've never seen such issues, actually. The first error is a very common one and harmless.
The second one can be anything, but one common reason are memory management issues, and most commonly having an object being destroyed twice - usually users destroy an object that will eventually be destroyed by XData.
But it shouldn't crash the server at all. It's hard to tell what's going on. Maybe you could try to use some stack trace tool like JclDebug or TMS MemInsight and log the call stack of the error too?