I thought most people were having the best luck scaling by a ratio of the diameters, not the cross-sectional area (though area makes more sense)?
For me, turns out area was more accurate, but ran me in to the ps_w limit sooner, which was causing problems.
Either way some underscaling (either global, in MLHFM, compensated for by KRKTE, or selective, in KFKHFM, compensated for by KFLF) is required.
Currently, I do a combination of both.