In fact in my opinion the importance of photon noise is overestimated. No picture gets ruined because of photon noise, everytime noise is a problem there is a read noise problem.
Ok, it may have been worth mentioning in your demo and previous posts as quoted to Bill that you were referring specifically and only to read noise differences in a disaster scenario when you suggested
"For the same amount of total collected light and DOF, a M4/3 sensor needs two extra aperture stops over a FF sensor, that is right. But in that case, noise statistics dictate that the M4/3 sensor will double the SNR over the FF sensor". That was not obvious to me. And of course it's not true when all else is equal/equivalent - in fact in practice it is mostly not true at all
In practice the impact of read noise in this type of comparison is limited for the typical photographer. For instance Bill Claff has a neat way of defining the ratio of maximum to minimum useful/acceptable signal in typical photography. He calls it Photographic Dynamic Range and has calculated this value and much more for a large number of cameras at his site
http://www.photonstophotos.net/. That clever metric is based on Bill's realization that most photographers will find image data below a certain SNR unacceptable when viewed in typical conditions. If interested in the subject I encourage folks to read up on its details there.
Using the earlier example of two cameras from roughly the same generation and number of pixels to guide us through this, PDR says that image data with an SNR below 5 would be unacceptable for photographers with 16MP cameras like the EM-1. His fine site also indicates that at ISO 3200 the EM-1 has a read noise of 1.4e- and the D610 at ISO12800 of 2.4e-*. The Signal at which the normalized D610 would hit a SNR of 5 is about 29.7e-, this is the minimum acceptable Signal. As a result of its lower read noise, theory says that at that Signal the EM-1 would instead produce a SNR of 5.28, not substantially different from the D610's.
Theory is confirmed by looking at actual images and measuring SNR off them, as I did in the earlier post**. The EM-1 at ISO3200 1/40s f/5.6 below received two stops more Exposure than the D610 at ISO12800 1/10s f/5.6:
Even in a disaster scenario, with the D610 limited to an SNR of 1, the EM-1 would have an SNR of 1.33 - the difference wholly accounted for by the EM-1's lower read noise and still very, very far from 2x. I would also like to point out that all formats are fast converging to the current read noise limit of just over 1e- per pixel in 'dark' ISOs (see FF cameras released over the last year or so), therefore it's unlikely we are going to see such big differences in disaster scenarios in the future.
Jack
*Normalized to 16mp, but normalization is not a determining factor for this calculation.
** From DPR dim studio scene, 'comp' button pressed, many non idealities and provisos apply.