The problem with that is you’re still bottlenecking the entire website/application.
One of the coolest features of SAFE is that it provides near infinite scalability for applications, because SAFE will offer simple key/store manipulations. This happens because applications SHOULD be built completely decentralized, as the network itself is.
That means that you’ll see a bunch of negatives caused by forcing decentralized SAFE clients to use the defunct server-client interface. For example, you’ll be stuck with modern scaling problems (you need more and more servers to keep up with demand). Additionally, the network overhead seen by the client will be too much compared to the speed of serving up an actually decentralized app.
It also seems more reasonable to just write your server-processing-application (let’s call it a video encoder) as a client side library, so the client is doing the work of encoding, and then verify their work (zkSnarks!).
I think there’s a lot of dependants on the use-case. But this is a new paradigm of application building that shouldn’t need a server to dynamically insert your username into the page if you’re logged. It’s archaic.
I answered some of that above, but essentially it’s a new way of serving up data, and you’re not requesting it from a specific server (as you would when watching Netflix). Instead, files are stored in multiple places, and the network is trying to serve them up to you as fast as it can. This includes intermediate caching by vaults.
Structured data is a ballgame I’m not fully prepared for, although this might help.
Distributed computations = I ask you to solve a really hard math problem. You solve it and send back the solution with a special verification key (that you created as you solved the math problem). I can then verify your solution against the verification key. I used only a tiny amount of processing power myself to merely verify your solution, while you did the bulk of the work actually solving the problem.