Standard warning: don't worry about twin operators unless you have unusual intensity statistics - the mere fact that your lattice is compatible with twinning does not mean that anything is wrong.  What does the final verdict in Xtriage say?  (By the way, the new version in the nightly builds provides much clearer feedback.)

Although it's possible that you have pseudosymmetry, I think this is ultimately the core problem - no matter what space group you try, the MR solution is incomplete, and probably for the same reason.  My guess is that the conformation of the search model is significantly different (either locally or globally) than that of the crystallized molecule.  If it's a homologous structure (not the same protein), try running Sculptor first to trim off the parts that don't agree.  If it's the same protein, either try cutting it into individual domains and searching for those, or trim off extended loops manually.  I would use P212121 symmetry if possible, but it doesn't hurt to try MR in all of the space groups for now.

If you want to send us the files separately to help@phenix-online.org, we may be able to give a more specific answer.

-Nat