Good question, can’t help thinking it’s kinda rhetorical. I’m rusty on the network layer, but would imagine it would be possible to come up with a usable algorithm to tell whether a packet contained text (HTML etc) or binary data (images, audio, vids). But given that copyright is a good few layers above, there’d be no way of spotting that – any kind of marker/pattern-matching system would have way too much overhead. The only alternative I can think of would be using knowledge of specific hosts – which seems to run counter to the neutrality angle.