That doesn't get away from the issue of payment, it just moves it around. I think it does so in a way with a lot of positives, but there's the whole P2P bootstrapping problem in the way, both in terms of creating the code that even works and then actually bringing the network online, and unfortunately it seems that's a pretty big negative in practice.
It just "moves it around" to the point where the cost is so spread out as to be effectively zero. What's the average cost of a volunteer-run peer to peer file sharing network?
How many computers running P2P software does it take to write the software?
You can only talk about the "all but 0" marginal cost when you can produce a piece of software that really does have an all-but-0 marginal cost. And, therefore, doesn't require the computer running it to be always on, or run on a dedicated VM, etc. Sure, such a thing may be possible, but you have to produce it first.
Unfortunately.
The cost comes in the labor of the organization that needs to bootstrap and maintain the network of volunteers - this includes marketing to them that it's safe and worthwhile to load software onto their computer. For instance, SETI@home has a budget in the hundreds of thousands of dollars per year: http://setiathome.berkeley.edu/sah_budget.php . Who would donate to Snapchat@home when it's just as easy to get Snapchat+ads on your phone?
See the Diaspora project for how great/easy P2P social networking is. Getting P2P working on bandwidth and port limited smartphone connections sounds like a nightmare.
No. It couldn't be a peer-to-peer setup because you couldn't guarantee that all peers would delete photos after they expire. This has to be centralized in order to guarantee that.
Well, you could have only one user (such as the sender) hold the decryption key and be willing to securely send it only to the recipient within a given expiry... The distributed data could still be floating around but would be unreadable.
There are plenty of distributed designs which could be secure and "read once within a time period".
(btw, a distributed network topology can look isomorphic to a centralized topology at the application layer. You seem to have a highly simplistic map of the territory.)
Isn't this the sort of thing that could be implemented as a peer-to-peer networking service?