Here are some suggestions:
1) For the offending residues, check the TLS groups. Maybe you can exclude these residues from the TLS groups (using isotropic Bs for these residues only) or come up with a better assignment.
2) From your description, it sounds like you refined XYZ, B, TLS and occupancies right after MR. You could refine XYZ and B first, adding TLS and occupancies later.
3) 1.5 Å resolution is typically the "border" for using individual anisotropic ADP's. So you might want to try using anisotropic ADPs instead of TLS. There is no guarantee that it'll work, but why not give it a try.
4) Unrelated to B or TLS: Check the distance between the oxygen of the peptide unit (which is pushed out of density) and the nitrogen from the symmetry related molecule. I circled the interaction in red in your figure. The distance should be around 2.7-3.5 Å. I cannot tell from looking at the screenshot. If it is shorter, nonbonded restraints could push the atoms apart.
Best wishes,
Dorothee