Think of variation selection as being more appropriate when what we are talking about are for most purposes simply *free variants* for presentation -- either is equally correct to most people under most circumstances -- but where for particular presentation purposes someone wishes to choose out a precise variant and have indication of that usage reside in the text stream itself. (And even then, this is only used in extreme circumstances when failure to have such a mechanism available is causing a mapping problem or similar issue which threatens to become a character *encoding* problem for the committees.)
This is *not* the case for the Mongolian FVS, by the way, one of the reasons that we didn't use generic Variation selectors for that script.
I'm not(!) advocating a Bengali FVS, but adding such a beast would in theory overcome Ken's objection about ignorability of variation selectors, as it could have documented behavior that's not generic.
However, that's got to be about the second least attractive option imaginable. (Leaving the slot for truly least attractive option open here for some as-yet-undiscovered monstrosity ;-)
A./