That is a good point about the system noise. However, when using data from Photonstophotos there is another factor that comes into play. From my understanding of the explanation of his methodology the results are normalized for a standard output size. There is specific mention of correcting for circle of confusion when calculating results for testing full frame sensors in APS-C crop mode. So not sure how much of the difference in his "photo dynamic range" numbers are measured vs calculated.
I believe measured, but the starting size is normalized so viewers would view equal image size side by side.