I liked Michael's article a lot. Very good.
Thanks to Michael for this analysis and for sharing with us his experience, and congratulations!
What this means is that a full frame sensor with, say, 12MP, will always have lower noise and thus inherently higher image quality than a full frame sensor with 21MP.
However, I cannot agree on that point. Noise isn't the only variable explaining perceived image quality. The question is more complex. Let me explain a bit.
1. "Preceived" Image Quality depends on many factors in a non-linear way: real detail, contrast (micro and macro), dynamic range, tonal variability, noise, light and subject properties, etc. and even a random variable.
2. Lenses have a lot to say about real detail finally captured. Given a size for the sensor, more pixels means a higher sampling frequency (better signal reproduction, as transmitted by the lens) and higher "reading" contrast (the sensors also have MTF functions). However, on the other hand, more megapixels means a bigger picture, this is, a bigger matrix that contains visual information. But the real amount of this visual information doesn't grow as fast as the image size (efficiency problem). Compare, for instance, the Canon 1Ds Mark II and the Canon 1Ds Mark III. The image size is much bigger (22 vs 17, 14 bits vs 12 bits), but the real detail differences aren't as big as that. This is due to decreasing contrast at higher frequencies in signal transmission by any lens. Summing up, more megapixels means more real detail, but there are disminishing returns.
3. Even if real detail and contrast improves with the number of pixels, there are other variables at play, dynamic range and tonal resolution or signal/noise ratio among others.
4. Therefore, it is true that a 12 MP camera can be better than a 22 MP camera in several points (dynamic range, tonal variability, signal-to-noise ratio) and not so inferior in others (real detail).
5. But... Better image quality can be easily perceived when all of these variables move together in the same direction. For instance, 22 megapixels from medium format versus 22 megapixels from 35mm format (at low ISOs). The MF image would be better in all respects. Or 12MP from 35mm versus 12Mp from APS-C at the same ISO value. Etc.
When several of these variables move in one direction and others in the opposite, the final result (image quality perceived) depends on other parameters not previously formalized (the subject, the output format and size, viewing conditions, etc.). Michael considers this when he talks about print size and resolution of a 12MP image, for instance.
