This is my original disagreement with Ed's message. It can be done, and when you do it it works, but it is too difficult for most people to bother with. I think we all agree on those three facts, just not on what to label the last one.
Actually, when I wrote "it does not actually work" I meant all three things:
1. It can't be done as a user would like to do it; note also that even experts
do it incorrectly (it's just too many detail devils).
2. When a user does it, the user does not really know if it was done right.
3. It is too difficult for users to use and (worse) most users who use it
do it incorrectly.
We have some choices. We can continue to say that it works and just wait
for users to get educated someday. Or, we can say that there is no x (x = market,
need, risk, point) -- and that's why no user bothers with it. Or, we can try
to understand what's it that users reject and work around it. My opinion I
already say upfront: users reject the whole model; it's not "natural" to
ask me for my envelope before you can send me a letter.
(btw, name and mail address are not the envelope -- they are routing
information. My public-key is the envelope analogue when comparing postal mail
with secure email.)