About “the origin server could supply this information”, it’s true, but there is a prevalent web philosophy about getting the content from the source if you have its URI, even if you are provided the content in parallel. It’s true of ActivityPub where retrieving an object from the source server, for a boost for example, is preferred to using the content possibly included in the boost message.

As well as having the most up-to-date content, it helps verifying the authenticity of the message.

There are certainly reasons speaking in favor of the current model. These don’t make it efficient however. If Fediverse continues to grow, it will eventually (and rather soon) reach a point where the current model simply won’t be feasible – and be it because it essentially becomes a DDoS attack. So at least limited centralization will become unavoidable here.

Note that “verifying the authenticity” isn’t quite possible. The server might not see the same thing a user would.

@lightone

I agree this isn't efficient.

And I'm not sure about the exact workings of ActivityPub, but there are ways to retrieve content from a server the way a remote user would see it by providing some form of authentication in the request itself. This is how it is possible to exchange private user content between remote servers.