Jeremy Kun recently wrote a good article summarizing some recent advances in encryption that make your statement somewhat less-than-entirely-accurate (scan for "differential privacy"):
http://jeremykun.com/2013/06/10/why-theoretical-computer-sci...
Yes, someone will inevitably lose both. You just need to ensure that that is a rare event, and that there are alternative systems in place (i.e. that losing access to one system does not prevent people from living their lives).
And there's the rub. "What do you mean, I can't ever see my data again? Why can't you reset my password?"
We know that true security means only the user has the key. But users don't all want that responsibility.
The catch there is that IBE requires a centralized, trusted key-issuing service where you need to enroll to receive your message. If that's compromised, then game over.
http://www.argreenhouse.com/society/wcan06/wcan06s4p3.pdf
Of course, you would need to be judicious about which group of key issuers you are willing to trust, but this method will at least reduce the risk. The other nice thing about this is that even if some key issuing service is compromised, the sender can force the receiver to switch services (compare to the TLS model, where dropping a CA is basically a coordination game problem).
Securely in the Browser, and not directly accessible by js, just like a TMP chip or a smartcard. Could that be a solution?
Of course, that introduces a new problem: securely backing up the private key.
As for using js for public key encryption, I've implemented it for a client and didn't have much trouble. There are libraries that workaround the usual problems. What have you seen out there that would cause a problem?
That said, couldn't this be mitigated by having a strong passphrase on the private key? How hard is the wrapper to attack?
Also, couldn't security researchers easily monitor the packets on this process and sound the alarm should they find that the js served up by Google or Microsoft suddenly starts sending private keys to the server?
As for your second question, there are techniques that perform static and dynamic analysis on javascript to try and detect illegal flows or taint propagation (without having to resort to monitoring the outbound network traffic). See [1] and [2] if you're interested in that topic.
[1] http://static.usenix.org/event/sec10/tech/full_papers/Bandha... [2] http://publik.tuwien.ac.at/files/pub-inf_5310.pdf
I'm not sure I agree that user-friendly cryptography is "intrinsically difficult." It doesn't seem like it would be hard for email clients and even the Gmail frontend to pop up a message saying, "Your email is insecure. To let people send you private messages securely, set up your 'public key' now. It's easy." Then a short wizard would walk users through the process and automatically append the public key to all outgoing messages.
On the other side, if you were going to send a message to a friend, the email client would check if that person has published a public key and then ask, "The recipient allows secure messages. Would you like us to send this message securely?"
Google and Microsoft and other large companies are no strangers to implementing a feature and using their size and clout to quickly make it a de facto standard. The real reason we don't have easy end-user cryptography is that these companies would lose access to mine your data and provide new services on top of that (and the article mentions this too.)