I agree but only to a point. I actually have a piece on my blog I will be submitting here today which discusses the idea of a PKI for the current threat model (resourceful organized crme and secret government surveillance. I have an idea for how to build a PKI to help combat these and laid my ideas out there.

But what is relevant for this discussion is that one of the goals has to be to increase the effort of a security compromise and reduce the value of one. Nothing is going to stop a government official spearphishing a target with an appropriate warrant, but if you can force the issue to that level of attack, and if you can make sure that the endpoints and only the endpoints are where compromises are vulnerable to sustained secret attacks, then maybe we have a chance. (It also means it is a lot easier to serve a warrant on the related computers rather than evesdropping silently in the middle.)