The difference today is this: lots of people are buying restricted computers (tablets, smartphones, even some laptops), and it would not be hard for the government to push backdoors into software distributed via the "app store model." This time there is actually a chance of the Justice Department getting their way.
So it will go like this - I write hi on the IM client, the Pi intercepts the protocol unpacks the hi encrypts it with the private key, sends some base64 to the server who then translates it to the receiver that has similar device installed which does the decryption.
It is not perfect but it can be done. And I am sure Mr Schneier can think of something much better in its sleep.
As a coauthor of the original post, I think the term is still reasonable and useful.
I agree that the applicability of "backdoor" to the cases you mention is less clear, and sometimes absent. Often, the main reasons that users don't get to control their own crypto keys may have to do with a service provider's business model and engineering decisions. If facilitating surveillance wasn't one of the main design goals, I can see where "backdoor" feels much less appropriate. (I think I'd argue that it's still an appropriate term if the architectural decisions about encryption are directly influenced by pressure from the government. Legislation to ban certain designs for communications systems should put us squarely in that category.)
I'd also like to find ways to encourage users to value end-to-end encryption, but I don't think that was meant to be the point of using the term "backdoor". We are actually talking about legislation here, supposedly meant to change how services are engineered, likely including services and software that are already in use.
Can't stop the signal, Mal.