Originally posted by JohnBee Samples updated,
Full size images added.
Thanks for the full size images (download links are reversed though). Also for the ISO3200 full size image which I would call exhibit C and add to the OP.
The 51k exhibit has lost quite a bit of detail, no wonder this is visible even at the reduced sizes you were posting originally. It is still a nice photo which would print nice, I guess up to A4. So, that's an impressive result for a ISO51k shot indeed.
Nevertheless, I'd like to see the 51k JPG prior to any NR (LR3 w/o NR and no Denoise) so we can see how much of the loss of detail is in the pixels and how much was eaten by NR.
The ISO 3200 full size is interesting. At first look, it seems to have more detail and more contrast than the ISO 200 one. That may indeed be the case given the difference in exposure times.
OTOH, I've seen occasions where higher ISO shots had better numerical resolution figures when analyzed. That can happen if the tone curve gets steeper because of running in a regime with less dynamic range headroom. E.g., ultimatively, a binary B&W image (2 levels) would have maximum resolution figures and higher ISO reduces the effective number of levels.
So, the entire topic is full of traps.
But that's a positive thing: the difference between ISO 100 and 3200 becomes so small (rather than obvious) that secondary effects start to become important. This is new