I've been thinking about this, and I feel that while #2 is definitely convenient in many ways, it really does set up the wrong mental model and incentives when it comes to the network.

To me, an important point indicator of the health of the ecosystem is that it should matter as little as possible to an app(view) whether a record came from them or from a different app that's writing the same lexicon. Once you start putting a AppView as a recommend middle layer, now it suddenly does matter which AppView you talk to. Sure IF all they're doing is proxying the request to the PDS and caching a little bit of info then that will probably work out, but I can't imagine this won't end up abused in the long run. 

Unless there's strong "to be a good ATizen it is all you should be doing" expectations from the dev/users community, you'll get apps trending toward doing what's convenient over what's right, especially as the ecosystem grows and attracts different players.