ROHF MO energies aren't uniquely defined quantities, unlike purely canonical RHF or UHF orbital energies. In the latter, you have one set of Hartree-Fock conditions (that first-order mixings between the occupied and unoccupied orbitals must be zero), and the canonicalization condition that the Fock matrix be diagonal is simply a more restrictive version thereof. In ROHF, you have three conditions that must be simultaneously satisfied: mixings must be zero between doubly occupied and singly occupied orbitals, between doubly occupied and unoccupied orbitals, and between singly occupied and unoccupied orbitals. Different quantum chemistry programs typically choose a definition of the diagonal blocks of the Fock matrix that is computationally convenient. (PSI, for example, at one point chose diagonal blocks that were an average between alpha- and beta-spin Fock operators because it led to simpler forms of the orbital Z-vector equations in configuration interaction gradient theory, but that's a long story.)
Should you "never trust MO energies from ROHF"? That depends on your purpose. Given that MO energies aren't observable quantities anyway, I wouldn't tend to "trust" them for pretty much anything. However, if you want to compare them to ionization energies or electron affinities, for example, just be aware that not all definitions for ROHF MOs yield expressions that are comparable to the corresponding Koopmans-like equations. This depends on the quantum chemistry code you're using.