Ah - cool. Let me know when you get that checked in (hopefully somewhat soonish?) so I can push a new version of libMesh to my users.
There's actually one more chunk necessary to get
LIBMESH_BEST_UNORDERED_SET<Node*> working: we need to define a
hash<Node*> specialization on any platforms that fall back on
__gnu_cxx::hash_set or they'll fail to compile.
I had a macro to do this for a PECOS chemistry library at one point;
I'll see if I can hunt down that code and pull it into libMesh now.