Updated patch accordingly. 
Regarding the problem in http://mail.python.org/pipermail/python-dev/2013-August/127862.html , it can indeed be solved by returning the already existing module object, if PyInit is called multiple times. I followed the discussion and can not make out a definite decision on how to handle this.
My understanding is, that up to now extension modules are only supposed to be initialized once per interpreter, so the check I have included in, for example issue15651, makes sense from this perspective. There have been reasonable desires (from Eli) to make the module state separate from each imported module within the interpreter, but this would involve more drastic changes and will be rather part of future Python releases.
Should we therefore (for now) make this a mandatory PEP3121 convention and include it into the xxmodule.c refactoring?