This portion of your post confuses me.
First, can you outline why CFA filters are not as good as interference filters? In what ways are they inferior? How does the mere presence of the CFA filters cause a problem? I do realize that many think CFA filters are very inefficient but that is mostly because of incorrect reading of the sensor plots vendors give us. See the following post for what I think is the right way to read the data we are given concerning CFA filter responses.
That may be true, but to recover saturation you have to apply a transformation to the signal that amplifies color noise. That's not a minor hindrance, and it effectively reduces SNR.
While you may be collecting lots of photons, not classifying them accurately is a form of error. As with deconvolution, recovering from that mixup amplifies noise as a tradeoff.
Regarding the statement "The CFA interferes optically with the signal.": can you explain to me how the presence of a CFA filter over one pixel introduces optical interference with adjacent pixels? And why would the same thing not happen on a pixel without a filter? At the pixel level, I cannot envision a process by which one pixel affects another just because of a filter covering it. Any detrimental effects would only seem to happen when you go to recover the color information for this pixel using its neighbors, to the best of my understanding. Using CFA Drizzle, we avoid that dependence on the neighbors to get a pixel's color information.
The CFA and the OLPF are a combo. The OLPF defocuses the image ever so slightly, reducing resolution. On purpose, to avoid color aliasing in the CFA. It can also cause other aberrations, like CA, although most high-end sensors I've seen are free from these defects.
Along the same lines of though, why would an overlap of filter responses cause loss of information. In my way of looking at it, the overlap enhances the information recovered. It actually aids the process by helping with the color fidelity of the data. I think of this as a case where "what CFA can lose in color resolution, it gains in color fidelity."
That's probably a contentious point. As I already mentioned, the matrix transform amplifies noise, and that decreases SNR. That's a fact I battle with every day. Does that completely undo the increased SNR from overlapping bands? Hard to tell.
Also, whether you want the overlap in astro imaging is a debatable thing. While it is true that given a monochromatic light source, overlapping filters retain spectral information, sources contributing a single pixel are rarely monochromatic. In order to recover that information you have to make assumptions about the source.
If you're going to make assumptions, you can also make useful assumptions with mono. Emission nebulae are narrowband, and usually emit Ha, Hb, SII, OIII. Interpreting nonoverlapping RGB into those bands isn't hard, even splitting Ha from SII may be possible, since Ha and Hb are usually correlated. Reflection nebulae and stars are broadband, which reflect/emit in the black body spectrum. Given RGB ratios it's not hard to extrapolate the full spectrum. In fact that's what PCC does.
My point is, that trick is not exclusive to OSC. It may just take a different form in mono.
An example of this might be to look at a particular emission line in a target. You can choose something like Oiii or Ha. With non-overlapping interference filters you capture the Ha line with the red filter only. With the CFA filters, you capture the same Ha line with a combination of red , green (~11%), and blue (~4%) filters. Now reconstruct the color via Drizzle for both sets of images. The interference filtered image only has data in the red channel. When displayed, the line will look whatever color of pure red your monitor can (or chooses to) produce. However, the CFA filtered image will present a color that is closer to how the eye sees the Ha emission line with a proper mixture of colors. The fidelity of the Ha reproduction is better because of the overlap of filter responses which more closely resemble the eye's response.
When I do Ha, I usually colorize the Ha channel to synthesize Hb. Similarly, if you have pure Ha data you can easily mix it in RGB to match your monitor gamut. Beware that depends a lot on your monitor and may not produce the same effect in other monitors.
I find that process a lot more understandable and flexible than depending on the overlapping responses of the CFA to do it on its own. That's of course an opinion.
Some day I would like to find the time to run an experiment as outlined in these other threads. I have both Mono and OSC cameras. I would like to take a bright light source and shine it through a prism. Then, the resulting rainbow would be photographed with both my Mono camera and AstroDon LRGB (i-Series) filters and again with my ASI294MC-Pro OSC camera. I strongly suspect that the rainbow simply cannot be reproduced cleanly (or pleasantly) with the Mono and filters because there just isn't enough overlap in the interference filter band-passes. The result would likely look very much "color posterized" compared to the result from the OSC camera. Without a good bit of PixelMath blending, non-overlapping interference filters cannot produce smooth tonal ranges. This just doesn't matter if you are going to be creating synthetic pallet images from the interference filtered images, however.
That very well may be true. I expect you'd be right there.
But we're not photographing rainbows are we?