A "SetMotherLogical(pMotherLogical)" call is missing in the G4PVDivision constructor. It should not be a big bug except that now, with the Geant4 8.0 release, it can let to a crash. In our application I verify that, as the flmother G4VPhysicalVolume's attribute is not set, the G4RegionStore::SetWorldVolume() takes it as another world volume. It lets to bad initialized physics tables and a crash when running G4VEmProcess::FindLambdaMax(). In this method, the call in the line 486, size_t n = theLambdaTable->length(); will returns zero and the following line, G4PhysicsVector* pv = (*theLambdaTable)[0]; lets to the crash (there is any test against n==0). I just patched the G4PVDivision constructor and now the G4VEmProcess::FindLambdaMax() doesn't crash anymore in our application. For information, we don't use for the moment any G4Region functionality. Anyway, to create a G4PVDivision should not let to a new world volume...
I placed at http://polywww.in2p3.fr/~mora/Geant4/ProblemReport829/ a copy of our patched G4PVDivision.cc which fix the bug.
The problem is now fixed for all G4PVDivision kinds. Thanks for reporting this. The fix is included in tag "geomdiv-V08-00-00" and will be considered for a future patch to release 8.0.