Imagine you own a webmail company. You secure your SMTP relays with SMTP-TLS (let's pretend it works well) and your client connections with TLS again.
Now your users are safe from random creepy flatmates, criminals with wifi sniffers, your telco and even mass government surveillance. But, governments can still serve a warrant on you to get email in a targeted manner, assuming they have a working MLAT process.
All you did was apply ordinary encryption to an ordinary website and you have that middle ground between "unbreakable to everyone" and "totally useless".
It's easy to confuse that with the name of a programme to reverse engineer and analyse Google internal traffic (which is presumably now over as all that traffic is encrypted), and I made that mistake myself, but I was corrected on HN no less by someone who is closer to the leaks than I am.
It wasn't something done with the knowledge of Google or Yahoo, as evidenced by the use of reverse engineering of the data formats.
But that's the point: we've seen that this balance doesn't exist, because if the service providers have access to the plaintext, the State will not contain itself to issuing case-by-case warrants - see Room 641A, "SSL added and removed here", etc.
The current move to encrypt everything is a reaction to the realization that the balance that was thought to exist, does not, and political participation is not sufficient because the information is kept hidden, so no informed political discourse is possible.
Is it? How? Because that's certainly not the default that you referred to.