Perhaps a similar concept as I am not sure Roger defines how to calculate unity gain.
DXOMark rates sensors to defined criteria. One of those is highest ISO without a loss of image quality. Per them 5D2 ISO is 1815 not 400. If Roger is saying its 400 there must be an error as there are many many fine 5D2 images taken at ISO3200
http://www.dxomark.com/index.php/Cam...EOS-5D-Mark-II
My understanding is when you increase ISO you reduce dynamic range. For example a lot of cameras maximum dynamic range is at its lowest ISO.
So if you shoot above this DXO Mark ISO you will get reduced dynamic range as the main consequence (blown highlights usually).
Theoretically you would get the same image using ISO6400 as at ISO1815 and then boosting the exposure compensation in software.
But I believe this unity gain concept is not 100% correct in that modern cameras also have clever noise reduction algorithims so you get better performance than you would have with earlier models at higher ISO than the rating may suggest.
If you were doing a tracked image where star elongation isn't the limiting factor to the exposure time (like it is with untracked images) then I would use the closest ISO to this DXOMark rating and make up for the lack of brightness with a longer exposure.
Again lens considerations complicate this further. A bright F2.8 or faster widefield lens with little or no aberrations (coma, distortions, chromatic aberration ) further enable you to get a bright image with minimal star elongation at a reasonable ISO.
ISO is merely an amplifier of the voltage before analogue to digital conversion. Like turning up the volume on your stereo. Eventually it gets distorted.
Greg.