C++11 is indeed a new language, but C++0x was being defined in public (And well, it's not like Herb Sutter would lack inside info either) and it was decently well known what it would contain years in advance. So it's not like the compiler team only could start implementing it last year.

Popular vendors taking many years to implement new language standards is a huge problem. It means fragmentation and lack of progress for years to come.

Why would they take the manpower and time to start working on things that may not be in the final spec?

For the good of their users? To not fall behind gcc? Anyway, specific features were finalized and accepted (or dropped, such as concepts) well before the final spec was released. The spec for varadic templates for instance was basically done in 2007 - five years ago.

To see if those things are implementable and useful. That knowledge then goes back into the spec process.

Sounds like this is something that's worth addressing on an architectural level.