I agree the security of this solution may be right for certain applications.
Yes, my proposal is an infrastructure play - but I believe it is a more realistically achievable one than most password alternatives.
There are of course some questions, that also would need to be solved. For example: What if an other person has access to my computer -- than I still would need some kind of master-password that had to be remember-able. Also there must be methods, to transfer my keys to an other computer or to "copy" for example onto mobile devices.
The example of the SIM-card for banking just shows, that the "big security" is really a hard problem. Two-factor is one idea, but it is always a game where the attackers are hunting the defenders (or vice versa).
I see the transferability as an export/import format issue. W3C, browser vendors and password managers need to agree on a format for keychain export/import.
Note that this technique could still be used in conjunction with multi-factor auth. Since it is just an evolution of passwords, those type of mechanisms are still relevant and compatible (though, eliminating knowledge factors should reduce need for a second factor).
So far, both the W3C and the browser vendors have been AWOL for the most part on real solutions to this problem.
If you are interested, I'm looking for folks to help flesh it out and promote it. Any and all help is welcome!