@JaggedSac
1) don't really know. I don't have access to the New Scientist article, and I only know the numbers from the article I linked. So I assumed it was an average scenario.
2) 4cm^3/2cm would mean 2cm^2 on the xy plane at far end of depth range, or 1.4cm linear error. Thus the resolution loss factor would be more 4.5 than 3. But it's still reasonable as it would mean having a 4-5 camera pixel margin of error on where the limb node is detected. Looks perfectly fine to me, as a wrist would be about 20-25 camera px wide and we're talking of fuzzy logic trying to pinpoint where the proper limb node is to be placed in that span.
Let's even say that 4cm^3 applies at top range, you still have a 0.7cm linear accuracy on xy plane at half that distance, ie you're still around 10 pixels of error on screen. And I've been very generous in using a meter wide space for pointing... to be comfortable you will probably only use about half of that, thus getting half the accuracy.
As I said, even assuming the best from the numbers we're told it sounds precise enough for GUIs, but certainly not as much as a wiimote for games requiring fast and precise pointing.