In a follow up comment on his WebFinger post Joe Gregorio gives some perspective on AtomPub and XML:
Look, AtomPub has this problem, and if I had to do it all over again I would build AtomPub in JSON. An implementer wants to go from bits on the wire to a native data structure they can interact with in their programming language. I used to think that was laziness, or lack of knowledge, but it’s purely pragmatic. By using XML you have introduced a layer of indirection, you’ve taken a data structure and converted it into a tree based document which then has to be converted back into a native data structure, but now that has to be done on a per language basis.
I’m in the lazy/pragmatic camp :-)
I really don’t want documents to describe data, I just want a nice way to serialize the data and right now my favored way of doing that is JSON.