>By the way, I understand that you do not recommend "symmetry" and you do not recommend "ROHF". But WHAT do you recommend?
I recommend C1 point group (to allow unsymetrical occupation of orbitals in "symmetry-identical atoms", UHF (to allow unpaired alpha and unpaired beta electrons to reside in widely disparate orbitals) and "manual" construction of initial guesses using the method of Szilagyi, R. K. and Winslow, M. A. (2006) J. Comput. Chem., 27: 1385–1397 .
You will still need a large amount of work and (especially) luck, though :-(