There will be a lot of resistance -- in fact, the idea does not advance, someone has to implement it first (perhaps one of the many "security enhanced" Linux versions out there, such as the NSA's, implements this or a suitable framework for implementing SIDs at the kernel level).
Of course, filesystems too need to support SIDs.
And the old flat UID/GID system has to remain available for backwards compatibility.
In other words, processes need to have multi-component, extensible credentials. And I say they need to have an array of creds, with each thread (or clone()ed process) having an "effective" cred array index.