The part of the grasshopper implementation that seems tricky to me, as explained to me by @DavidRutten, is that in the typical grasshopper workflow every transformation creates an entirely new object. This seems like it could become a massive memory hog if one was transforming a subD via moving subobjects with named selections. This is why I phrased what I suggested so I was describing more of a remote control rig than an inside-grasshopper transformation process. Presumably, because of the way subobjects work in named selections, this would mean that one could only manipulate them via a remote control rig, but not delete or add subobjects.
Here’s that prior discussion with David: Sub-object data or referenced data in GH2