Hi,
I would suggest that large pixels have small advantage in DR, compared to downsampling. So if we need a 100000 ISO camera, largish pixels make some sense. On the other hand, I don't really think 100000 ISO makes any sense.
The way I see it, a modern sensor has a base ISO around 100 - 200. If it has very large full well capacity it may be closer to 100 ISO. So going to high ISOs it just means that we only use a small part of that well capacity, and no PR department can go around that fact.
With low utilisation of full well capacity we are going to get a lot of shot noise, due to photon statistics. If we reduce exposure enough we are going to see some readout noise. That readout noise can be reduced by using large pixels. If we combine four large pixels into one we will get four times the full well capacity., but readout noise would stay the same with all other factors constant.
Combining four pixels in software would still give us the same full well capacity, and therefore also the same shot noise. Readout noise would add in quadrature, AFAIK, so it would be doubled. So there would be a benefit in deep shadows but not much else.
Now, if we get back to DxO, all their data is based on noise, with SMI being the only exception but SMI is not included in the figure of merit.
So, I would say that DxO makes a lot of noise about noise. That said noise is an important parameter.
Another development is decrease of pixel pitch and increase of fill factor. Both are important for correct rendition, but correct rendition seems to be low on the priority list of photographers. Camera makers have a bit other priorities,but making customers happy is a high priority.
Best regards
Erik
Down sampling in post can provide most, if not all, of the noise reduction obtainable by increasing pixel pitch, and there are many other advantages of low pitches:
http://blog.kasson.com/?p=6078
I believe that the reasons that Nikon kept the resolution of the D5 down include:
20 MP is already overkill for most of the target market.
Frame rate is important to a significant segment of the target market.
Post processing speed is a concern to people working in the field on laptops.
Jim