People capture 21 MP all the time and use less for all kinds of needs, there's nothing nonsensical about it.
Nonsense refers to foolish 'rational' of comparing 2 products using one at a reduced quality. Obviously cameras will often have their results viewed at varying and much lower resolutions than capture, such as on the web, but that is not relevant if trying to compare and contrast the best
performance of the two items.
People don't buy cameras solely based on their native capture resolution either, although it is a very important consideration.
Actually that apparently is exactly
what they do and why MP are used to sell cameras.
Show two lots of people 2 sets of prints. One set a better quality than the other and most unsurprisingly will choose the better images. But if you tell the people that the poorer quality shots were taken on a camera with more MP [before deciding] and suddenly they will tend to prefer the inferior images.
I was going to start a topic on this rather interesting finding, but it seems germane to mention here.
If the idea is to compare the quality of the capture of two camera systems and their respective noise, one is 12MP the other 21MP, its never going to be a totally complete apples to apples comparison as I said. But sampling the 21MP down to 12MP at least tells you want the potential output of both, with respect to noise would be ASSUMING the output at that 12MP is the max. Are you proposing that instead, the 12MP should be sampled UP to 21MP? That would be far more nonsensical IMHO.
If comparing two products on their best
performance, you test in a way that is challenging, not one that is limiting. Unless you are actively trying to favour one product, which by limiting the higher MP camera to that of the lower MP camera is what you are doing. If only ever outputting to a max of 12 MP, why even buy/use the 21mp camera.
So producing large prints to the max ability of the 21mp camera, which is rationally why you would tend to buy a higher res camera over a lower res one [assuming it's not for bragging rights], then most people will upsample the lower res one to be able to match output resolution. So it's not so nonsensical, more representative of how large prints will actually be done. If you don't upres, then the noise/grain of the lower res file will be larger compared to the higher res file, plus detail willl be lower.
Surprisingly - upsampling does not always reduce quality. I tested the Genuine Fractals versus upresing in ACR some years back with the 20D [ACR was much better and free]. But what was surprising was that the ACR file that was upressed, looked better than the native sized file.