Currently the mmCIF parser can't parse phenix-produced files. The test added in commit 9cae80f demonstrates the issue (The Test tag is commented out for now in order for the build to work. It should be uncommented once this is fixed)

Looking at it quickly I think that the problem comes from the separation between loop fields. In PDB-deposited mmCIF files a line starting with a "#" is used, in the phenix-produced files a blank line is used instead. In fact the mmCIF specification doesn't even say that the "#" are needed. Lines starting with hashes are simply comment lines.

See http://mmcif.wwpdb.org/docs/tutorials/mechanics/pdbx-mmcif-syntax.html