@gargron consider I'm a third party that is interested in the activities of a user. If the stream is archivable as Atom, I can download the user's entire feed by following the rel=next links and thus understand what this user has liked etc. etc.
If I have to make multiple extra API calls to see the full picture, it becomes arbitrary and incompatible and instead of using an Atom parser, I'll have to write a bunch of intelligence to retrieve stuff from multiple endpoints.

Also: how do I callback-verify that something has been favorited if I can't refer to the fave activity by a URI? (playing catchup after downtime suddenly gets a lot worse)