I've found (as a rule of thumb) that, when asking a grad student "How much time do you think you have left before you can write up your thesis?", if the answer is two or more years out then it really means "I don't know." The student honestly believes this answer, but in reality he/she doesn't know how much he/she doesn't know.
I'm starting to feel about the same with Mir and Canonical here. Shuttleworth is the tenured but aloof professor who casually coaxes his students (employees) toward completing milestones but without too much urgency. Money's not plentiful, but the professor has enough contacts and contracts to keep his lab going and give a stipend to his students. They put out a few papers (releases) each year, and each time the students think this grand project is "almost done"... only to discover that there's still more left to do.
There's tremendous value in this kind of exploratory research. I'm just not sure it makes sense to package it up for end users.
If I were Mark Shuttleworth's technical advisor, I'd suggest examining RedHat's Fedora model. Create a small group called Canonical Labs where stuff like Mir and Unity can flourish, with continuous releases and without the artificial constraint of a set release date. (If this makes the environment too lackadaisical and development isn't progressing fast enough, find some other way to instill discipline and/or motivation; don't make it the threat of moving alpha code to end-users.) When it's stabilized (no longer shuffling menus and window icons around, for example), then integrate it with the main Ubuntu branch. Something a bit more edgy and up-to-date than Debian Stable or RHEL, but not so much that it constantly upends your users.