I can't see any reason why this wouldn't work, but you're right, it's
running untwinned ML refinement despite the twin law being set
correctly.  However, I just tried running the twinned example included
in Phenix (the porin-twin dataset, which can be set up as a tutorial
from the GUI), and it worked as expected.  I'm downloading the latest
installer now to double-check, but I doubt it's much different.  One
possibility is that the REMARK section in the input PDB file is giving
conflicting instructions.  Could you please run this command and send
me the output (off-list, unless you're certain there isn't anything
secret in there)?

grep "REMARK" /Users/nick/k94e_208/Refine_30M/ligand_revised2-coot-0.pdb

thanks,
Nat