tl;dr: how about a virtual global flat LAN that maps static IPs to onion addresses?
We all know the story. Random feature gets unintentionally picked up as the main reason for buying/using a certain product, despite the creator's intention being different or more general. (PC: spreadsheets; Internet: porn; smartphones: messaging.)
Exhibit 1: Ricochet IM (https://ricochet.im) uses onion addresses (each client runs a hidden service) as a sort of static anonymous IP address and, because it's static, it's the user's identity too, in a p2p/serverless chat app. It's dead simple, works like a charm behind the firewall at work, and protects metadata, which no other chat app/protocol I know does.
Exhibit 2: OnionShare (https://onionshare.org/) does the same for file sharing, and it's actually a much easier user experience to send large files this way than any other. Why? "Static anomyous IP" (onion address) and NAT traversal because all tor hidden services work by making outgoing connections to Tor relays and don't need any open ports.
Those are two great apps that, unlike Tor Browser (which I love very much, but hear me out), improve the user experience, through Tor, in comparison with the mainstream (OnionShare even more so). The user might not even care about security or anonymity, it's just a better experience, period.
You don't have to convince people to make sacrifices in the name of privacy, you just have to show them something they want.
That's when natural demands kicks in and suddenly you're not pushing water uphill anymore, you've changed the landscape and it flows in the direction you want. Like when Tesla made electric cars that people buy despite being electric, not because of it.
As good as Ricochet and OnionShare are, they still had to go through the trouble of integrating hidden services themselves.
If there is a virtual network interface that transparently maps static IPs to onion addresses, all sorts of things could benefit from the backward compatibility (old games, IP-based voip, screensharing, real-time collaborative writing, etc.) and new ones could be built a lot more easily.
[ZeroTierOne (http://redecentralize.org/interviews/2013/07/30/02-adam-zerotierone.html) does this, but doesn't worry about privacy.]
The only thing better than serving the privacy-conscious is serving privacy to those who don't even know they want it.
Of course massive use would probably crush the current network, but uptake would be gradual, and I imagine demand has a greater power to drive capacity than the other way around.
I'm nowhere near an expert and I could be just talking out of my ass, so please let me know if this is completely stupid and would never work. Thanks!
P.S.: you can follow me @obvio171 on Twitter.
P.S.2: more comments on Hacker News: https://news.ycombinator.com/item?id=8018213.
P.S.3: more comments on tor-talk: https://lists.torproject.org/pipermail/tor-talk/2014-July/033815.html (OnionCat seems to already implement something like what I describe.)