, is it likely that the subjective quality of images at large ISO may be different due to read-out noise being larger for Canon (but overpowered by shot-noise in simple DR measurements)?
An excellent thought: a sensor with photosites of higher well capacity but higher dark noise levels in electron counts can have lower engineering dynamic range, due to worse SNR in very dimly lit parts of the scene, but also have a better SNR in better lit parts of the scene, where photon shot noise dominates, and somtue SNR is the square root of the photo-electon count.
Indeed, in my rough comparison of two hypothetical sensors
A. one with well capacity 50,000 and no read noise at all
B. one with well capacity of 100,000 and read noise of 10e-
the former has greater DR (50,000:1 vs 10,000:1?)
But the latter a better SRN at any photosite getting within eight stops of full exposure.
For example, eight stops down, so with signals of about 100 and 200 respectively, sensor A has a SNR of 10, while sensor B has a SNR of 200/sqrt(300) or about 11.5, so the cross-over point is a bit more than eight stops down from maximum. This suggests that the sensor of lesser DR will if anything look better for noise effects on any straight print done with correct exposure at base ISO speed, because anything more than about seven stops darker than full well capacity is printed totally black, making the noise levels the invisible.
Of course, this is just yet another argument for using more refined measures of photographically relevant noise effects, not ”stops above the level where SNR=1". One idea, somewhatvendorsed by the ISO standard 12232, is measuring stops down to SNR levels of 40:1 (for excellent IQ) and 10:1 (for barely acceptable). On such measures, the bigger, noisier photosites of sensor B win slightly.
P. S. I originally constructed an example like this to illustrate why a medium format back with the current read noise disadvantage of CCDs can still in a range of practical situations have less visible noise, finer tonal gradations and such than even a "perfect" sensor in a format like 36x24mm, a bit less half the size and about half the electron well capacity (at equal pixel count) You just need to give the bigger sensor enough light to make use of its greater photon counting capacity, and avoid extemes of contrast compression in printing.
