Let me see if I understand the argument you've made: the idea that there are no layering violations is silly because people can make bad designs? Should we blame poor layering discipline for all the C code out there that calls gets(3)?

Modularity is a good thing. Composability often is too. For systems deployed in the large, the end-to-end argument also guides designs to simple cores and complex clients.

But none of that means that any given layer "owns" any piece of functionality. It would indeed make a whole lot of sense to start thinking of IP as the new Ethernet, for instance. But we can't really do that, because the greybeard priesthood uses concepts like "layering violation" to shut down discussion and maintain intellectual turf.