It turned out that the costs of separating the XSD datatypes are too high and probably not worth the effort, since with its limited scope probably nobody would want to use XSD.ex outside of the RDF.ex context anyway.
