OK, I finally just tested this for myself on my 5DII. Kept aperture the same and changed SS between 2s and 1/2s and my conclusion is......
I would test your own camera and raw converter.
In my test ISO160 was indeed less noisy than ISO100 in the deepest shadows. But in the mid shadows it was more noisy and ISO100 was the best. (seems like perhaps ISO200 with a tone curve applied ?)
In no way was ISO320 better than 100 or 160.
ISO160 had less 'dynamic range' than ISO100 (tested on the lightest part of the image, so not considering S/N of deepest shadows) and I couldn't tell the difference in highlights between ISO160 and ISO200 (which would suggest not just ISO200 reduced in exposure, otherwise I would have blown out the highlights a bit more with ISO160)
The same relationship existed between ISO320 and ISO200 - 320 better in the deepest shadows, 200 better elsewhere.
ISO100 seemed to give me more vertical striping than ISO160.
I've tested ISO50 before and the raw file is actually exactly the same as ISO100 - an exposure curve is applied in the .jpeg conversion in DPP (and I think in ACR). So the graph above is exactly what I'd expect for the shadows. The relationship between ISO100 and 160 (and 200 and 320) however does not show the same characteristics as between 100 and 50.
Personally the differences are not large enough for me to use 1/3 stop ISOs simply because I prefer to have 1 stop increments. I might try ISO160 for landscapes in the future though. Even if I do get a tad less DR, I do sometimes see banding in skies after increasing contrast significantly.
I wouldn't use ISOs 125, 250 etc.. though, they were definitely worse. Anything else, I don't think you can go wrong with using either 100,200,400 or 160, 320, 640.
Of course that's with the 5DII. I tested on ACR and Bibble as raw converters and although the outputs were slightly different, the conclusions held. The newest ACR is pretty clever at handling highlights, so that might be a reason to favor ISOs 160, 320 etc..
I can see why there is so much confusion actually and why the Magic Lantern team even has some difficulties in coming to a indicative conclusion on what is going on.
For video I would stay at the 160, 320 etc.. because you really don't want to ETTR or get close to blowing out highlights.