Does it have to be that way? Excuse my lack of fediverse and general networking knowledge, but couldn't a user instead retrieve the data directly from the instance a post is hosted on without their instance needing to interact with it at all?

I see how there's limited scalability in the example you provide, but I feel like it shouldn't be necessary for each instance to make their own copy of things and rather have the user get the data directly from the host instance.