Replies are (or at least can be) split into separate objects already - this was implemented ages ago, although I think @benwerd disabled it unless a config option was set, not sure why. I've been running split replies for years now without problems, so it seems reasonable to bring this back.
Splitting media to audio / video also makes sense, possibly with a back port.
So, really this becomes a search problem. I'd prefer not to have specific methods added unless there's no avoiding it, but if I remember correctly we already save the "indieweb type" with saved data, and if we don't we could without too much trouble - we already have this information available via ContentType.
If this was to be done on a standalone service I'd probably map this via a custom Elastic Search domain - done this sort of complex searching over objects before and it avoids a lot of the database headaches and scheme updates that'd be otherwise necessary. But for this, modifying the endpoint would be sufficient...