Unfortunately, XMLUnit turns out not to have a very useful diff; I had hoped it would do a real semantic diff. So it's largely a dead end.
Semantic diff of XML is actually a continuing research topic, e.g.
https://dl.acm.org/citation.cfm?id=2554893I found some ancient abandonware XML semantic diff libraries out there, and am looking into whether it might be worth my rescuing one and using it as the basis of something to plug into my test process. If not, I may just for my own purposes use a custom XML "normalizing" pretty-printing hack to put stuff on single lines in order to diff as a sequence of lines.