That's only true for NVidia and that's because they don't want to put the work in (They do plan to make it work on Linux though just not on pre WDDM2 Windows).
There is absolutely nothing locking multiGPU support to windows 10 other than the willingness of IHVs to support it.
Valve offered to donate the name OpenVR if it was needed, but it was decided that since the API will have uses outside of VR (i.e. AR and IR) the more generic name was more appropriate.
It's an intentional branding effort undertaken by Magic Leap to associate 6DOF HMD AR with the term MR instead of AR - and thus associate only themselves with the term like "Kleenex." Typical propaganda effort. Apparently it worked because Wired picked up the term and others like Robert Scoble did as well. FWIW they failed previously by trying it with "Cinematic Reality," but nobody bit.
The theory is that the term Augmented Reality has too much associated baggage accumulated over the years from implementations that don't do realistic 6DOF.
So they are trying to say - approximately - that anything that doesn't have depth is AR and anything that does is MR.
Graeme Devine made this distinction by throwing shade at the iterations of mobile AR from previous generations that didn't have SLAM or any kind of tracking in a recent talk [1].
In reality MR was defined by Paul Milgram in 1994, and does well to describe the spectrum that encompasses Augmented Reality, Augmented Virtuality and Virtual Reality.
"IR" - for "input reality" - might be a sensible alternative term for this. VR and AR seem to focus on the display aspects, while the likes of Magic Leap are useful in the input aspects (which have their own sets of problems).
