Believe it's just a cloud based mask rather than masking the Density PF and the cloud layer reading that as a single cloud. Taking into consideration cloud properties I guess.
As you can see, there isn't much difference in masking the cloud density, and using the density modulator. It seems the modulator method, again, uses some internal methods to determining the best effect for the mask. Masking the Density itself seems to yield a sharper, more exacting result, using the modulator seems to have some adaptive effects.