by 'multicast' here one really means a facility that's provided by layer 3. So yes, we can build our own multicast overlays. But a generic facility had two big benefits. One is that the spanning distribution tree can be built with a knowledge of the actual topology, and copies can be made in the backbone where they belong (copies in the overlay often mean that the data can traverse the same link more than once).
The other big one is access. If we call agree on multicast semantics and addressing, and its built into everyone operating system, then we can all use that as a equal access facility to effectively publish to everyone, not just people who happen to be part of this particular club and are running this particular flavor of multicast.