The person in the chimera-users post was going to use it for map-to-model FSC. Tom found that there was a pretty large effect in using molmap with defaults and then vop resample as opposed to molmap with the right box options, so he added molmap ongrid to make it easier to set the box.
I think you’re right about the use of molmap in general though. In particular it should never be used for map-to-model FSCs! You probably know, but for any one else that happens by the right way to simulate a map is with phenix.fmodel.
# Example with boxsize 256 @ 1.02 A/px.
sed -i '/^CRYST1.*/c\CRYST1 261.12 261.12 261.12 90.00 90.00 90.00 P1 1' 6bg.pdb
phenix.fmodel 6bgj.pdb high_resolution=2.04 scattering_table=electron
phenix.mtz2map 6bgj.pdb.mtz include_fmodel=True d_min=2.04 grid_resolution_factor=0.5