Company that was selling an "affordable" mocap (YEI) jumped on the VR bandwagon and suddenly instead of $4000 they just want $350.
How is that even possible?

They just have accelerometers and other sensors in the small devices he's got strapped to himself. The sensors' data gets sent to the computer, which then just applies the measured movements to the model. I saw either this product or someone with the exact same product at GDC this year showing it off and selling it. It's real. It works. One thing with mocap is that it's never 100% perfect. There's always jitter and stuff. Usually what will happen is someone will use mocap to get the major motions down of an animation, then have an animator go in and fine tune the animation so it looks just right.

I was going to buy two kinects then Microsoft showed greatly improved Kinect 2.0 and now even traditional mocap is going mainstream. Should I wait or just get kinects 1.0?

Edit: read my update below. Watch out. You probably can't do multiple Kinects looking at the same thing. The Kinect 1.0 uses structured light, and the dot pattern from one Kinnect sensor would most likely mess with the dot pattern of the second Kinect sensor. Interestingly, this company seems to be able to capture a scene using two Kinect sensors, and I have no idea how the Kinects aren't interfering with each other. I just sent them an email.

Kinect 2.0 uses time-of-flight, and should conflict with each other if there are multiple cameras.

Just something to be aware of.

Edit: Here's a YouTube video showing multiple Kinect interference. I'm surprise the interference is minimal (still there, but I expected worse). After some more googling, I found this paper by MS about minimizing interference of multiple structured light cameras. It's actually kind of impressive.

As for the Kinect 2.0 and time-of-flight interference... I guess we'll just have to wait and see when it comes out if they were able to minimize multi-camera interference. The only two ways of minimizing interference I know of here is using different RF frequencies (which would be quite hard when mass producing devices) or synchronizing RF emission (still hard, and not 100% perfect, but better than nothing).