So, what does expanded ISO range really mean?
There is no uniform answer, it depends on the camera. Let's restrict this to DSLRs, as MFDBs exhibit very different characteristics, and P&S? Well, who knows that.
As the first step, one needs to separate between low and high ISO extentions.
If the wells (the electron collecting buckets) get saturated, then any further exposure is either lost (the good case) or there will be a "spill over" to other wells; see the discussions about the D3/D700 banding. Thus reducing the ISO lower than the "native ISO" is
plain overexposure.
Thus shooting with the 5D2 @ ISO 50 will certainly yield very low noise even in extremely underexposed areas, but the dynamic range will be less than the maximum, due to saturation of the wells, which can not be countered any way except by lowering the exposure.
The "expansion ISO" at the high end is a completely different issue.
First, let's see what "real ISO gain" means,
in terms of digital photography, as opposed to the discussion in terms of electronics. You meter the exposure and make a shot @ ISO 100, with, let's say f/2.8, 1/125s. However, the 1/125s is too long, you need to reduce it to prevent motion blur. You have the choice between underexposing by one stop at 1/250s and shooting with ISO 200 at 1/250s. Then you make a shot @ ISO 200, with half the exposure.
What difference do you expect between
a. the original, higher exposure at ISO 100,
b. the plain underexposure with increasing the intensity in post processing,
c. the lower exposure with increased ISO?
Obviously, the highlights should not look differently, but how will the shadows look like?
With the plain underexposure one loses exactly one stop; that's obvious. But how much does the ISO increase help?
Example: the Canon 40D.
Increasing from ISO 100 to 200 is almost painless, i.e. the one stop lower exposure is almost unnoticable, it makes perhaps 1/6 EV.
From ISO 200 to 400: the loss compated to the higher exposure is about 0.6 EV; viewd from the other side: the gain is about 0.4 EV, compared to the plain underexposure.
From ISO 400 to 800: the gain is about 0.25 EV
From ISO 800 to 1600: the gain is perhaps 0.15 EV.
And now the next stage, from ISO 1600 to 3200: the gain would be negligable. Thus Canon decided, that there is no point to complicate the hardware with further increasing the nominal gain, when it does not yield any better result. Instead, the firmware "pumps up" the ISO 1600 values by plain multiplication, as if there were an ISO 3200 hardware setting, but the effective gain is nothing. In fact, there is a huge loss: doubling the pixel values means, that one stop from the highlights is cut off (assumed there is something in the highlights). Though this is the case with each ISO increase, in this case there is nothing in return.
Well,
that is the ISO expansion.