It seems, from looking at he published DR graphs of digital cameras (have not seen one from this B&W sensor yet), that there really is only one "real ISO speed" and high ISOs are manufactured by underexposure and sacrificing DR.
There is a lot of ambiguity in the phrase "ISO speed", which is why I try to refer to the multiple different measures of sensitivity and exposure index in the actual ISO standard 12232.
With film, the familiar speed measure was roughly the
maximum exposure index (minimum amount of light delivered to the film) that gave adequate shadow handling (loosely: noise floor four stops below the mid-tones).
The closest counterpart to that in the ISO standard for electonic sensors is the two noise based measures of
maximum usable exposure index, S40 and S10. These are, roughly, the exposure index settings at which the Signal to Noise Ratio in the midtones is 40:1 and 10:1 respectively. Looking at the DXO graphs of SNR 18%, those measures are up in the thousands for good modern sensors. [See note added below.]
However, those measures are rarely used: instead most attention goes on an almost opposite measure: the saturation based base sensitivity, which is the
minimum recommended exposure index below which there is inadequate highlight headroom. With modern sensors using microlenses and color filter arrays, this is in the range about 100 to 200.
I would like to see the sensitivity of a sensor decribed in the way that the ISO standard recommends: giving a range from a minimum (highlight saturation based) to a maximum (noise based, like S40). Reducing a complex mixture of sensor characteristics and performance goals to a single sensitivity number can be misleading.
P. S. The DXO graphs of SNR 18% in the "screen" mode (per pixel) give a rough reading of the ISO noise based upper limits on exposure index. Those graphs in fact have a red line at 20dB, which is a SNR of 10:1, considered as "barely acceptable" and in fact quite visibly noisy. The stricter SNR 40:1 standard is a common guideline for excellent noise levels, and corresponds to 26dB. Looking at some of the current state of the art sensors for noise levels, the 5D3, D800 and D4, I estimate ISO S40 maximum recommended exposure index values of
D800: 1600
5D3: 2500
D4: 3200
Of course, if one is satisfied with downsizing to DXO's "print" normalised 8MP , as is probably fine for most normal prints sizes from these low light extremes, the SNR goes up. The wonderfully simple result is that the ISO S40 limits become about 6400 for all three sensors.
This equality is, I believe, because in that comparison it is the fundamentals of photon shot noise at work. To partially confirm that, photon shot noise alone should give a 4/3" sensor an S40 speed in that 8MP normalisation that is reduced in proportion to sensor area (to get equal photon count per down-sampled pixel) so 1600: and that is exactly what I see for the two best 4/3" sensors tested at DXO so far, those of the G3 and GH2. The same argument for Canon's 1.6x smaller EF-S format suggests scaling down by (1/1.6)^2, to an ISO S40 speed of 2500 --- and again, this is what the DXO measurements show for the 7D.
