Maybe I wasn't clear enough! I NEVER spoke about improvement of BOTH parameters but having the option of large files with lots of detail and good color plus lots of calculations for debayering OR smaller files with less detail and superb color with much less need for calculation.
You were clear. And I'll repeat: there's not much left to be gained, beyond Bayer, in terms of color accuracy or fine color detail. Go compare images in the Imaging Resource Comparometer. Sigma claims 29 'effective' MP for the SD Quattro. The 5D IV is roughly equal in the color fabric area of the IR studio scene. (The 5D IV image needs some sharpening, but the data is there.) Switch to the 7D mk II which is closer to the physical or spatial resolution of the Quattro. The Foveon sensor is better in the color fabrics (except red which is a mess). Now switch to the 5Ds or 5Dsr and watch the Quattro get crushed in all respects.
IR's tool has the drawback of being based off JPEGs. I know ACR would improve the color detail rendition of all the Bayer sensors, I've seen it in resolution testing using a color map as the target. I have little doubt that the 5D IV RAW file would out perform the Foveon file, and the 6D2 would be darn close if not equal in all respects.
You can choose RAW in DPReview's tool but unfortunately they don't have a section which clearly stresses color detail the way the fabrics do in the IR studio scene. Which reminds us that we're talking about improvements at the very edge of details primarily defined by color (rather than luminance) variations. Coarser details aren't going to show the difference.
So we're looking at 33-50% more pixels in a Bayer sensor to match or exceed a Foveon sensor in terms of color detail rendition at the very edge of the sensor's capabilities...stuff you'll only notice while pixel peeping or inspecting very large prints. If you take a 100mp Bayer sensor and spit out 25mp images you've
lost color detail.
About 8 neighboring pixels: They just aren't drawn because I had to fiddle a lot to make just that graphical representation because fixed fonts seem not to be fixed enough
You're missing the point that modern demosaicing algorithms are already recovering more color information than most people assume. There's a persistent myth that we're losing something special or important by using Bayer. That would only be true if Bayer had the same spatial resolution and ISO characteristics as a stacked sensor. Bayer has raced passed stacked sensors. A 50mp 5Ds does not lack for fine color detail at any print size you or I are likely to ever make. What will a 100mp sensor lack?
At least one point about debayering: It doesn't work well with monochromatic light sources like e.g. LED tail lights of cars or traffic signals - these show strong artifacts which look like a chess pattern of R_R_R and the next line _R_R_ which is - just my idea - of the missing brightness data in the G and B subpixels: Pure red means that G and B subpixels see close to nothing. Downsampling in camera with quad pixel binning avoids this effekt.
I've actually tried to isolate and exploit this in the past. Everything I tried suggests that artifacts on strong neon or LED light sources are much more about channel blowout. The only way I could create the situation you describe (only 1/3 or 1/2 of the pixels responding to the scene) is through narrow filters on the lens.
And if you have that situation with a 100mp sensor then downsizing in post is little different from downsizing in camera.