Recent research has had controlled landing on 2 motors in real robots (and one motor in simulation only). The main trick is to spin the whole vehicle around the z axis (up-down) very fast, and modulate the speed of the remaining rotors to apply forces at different places around the centre of gravity. IIRC the one-rotor version didn't work on a real robot due to limits on the frequency response of the real-world motor controller.

Spinning the drone so fast that the modulation is usable doesn't itself sound very safe to me.

It's an upgrade from "expensive rock with sharp blades falling out of the sky" to "expensive rock with sharp blades falling out of the sky spinning at X0 RPM, and the sharp blades are also spinning at X00 RPM"

Sounds plausible but I haven't read about this before and would like to know more, do you have a reference handy?

In regards to turning on the z-axis to be able to land with engine failure, I believe User is referring to Autorotation: http://en.wikipedia.org/wiki/Autorotation

I thought autorotation was only useful if you're moving ahead very fast (high ground speed) while also falling. Is that right? Most drones are essentially hovering much of the time.

No, it's not autorotation. Here's a lovely demo:

https://www.youtube.com/watch?v=ek0FrCaogcs&feature=youtu.be

Patent-pending.

Yes!! Much better than a parachute, that needs altitude to work.