As I wrote in the initial post there is a difference between the two formats, 3FR = uncalibrated raw data with calibration data in tags on the side, and FFF = calibration applied and calibration tags removed. A converter that parses the format gets to the same result of course so from a user perspective there is no difference. However it's relevant to us that write raw conversion software. Also worth noting is that many smaller third party apps have its raw conversion based on dcraw, which does not apply calibration data as the format of it is not reverse-engineered. A while ago I reverse-engineered that for the IIQ format (now available in latest dcraw), and obviouslyt I have more work to do now with 3FR. I don't like to store my images in proprietary formats, so the least thing I want is to have the format reverse-engineered and covered in open-source software.
MFD CCDs are harder to convert than newer CMOS, as the former is much more "analog", there's slight non-linearity in the AD-conversion (hence calibration curves), and there's more read noise which can disturb demosaicers and create false texture.
Even the calibrated FFF file for my H4D-50 has slight difference between green channels, which put some demosaicers off target and they start producing mazing artifacts. This can be compensated for with a threshold though ("green equiliberation" in RawTherapee where this is a user setting), so its quite easy but still an extra thing to think about.
I do not agree with the view that it's about to "extract" the DR they capture, you can't really do magic there in signal terms, the raw data is what it is, but you can do "psychovisual magic" ie, make a noise reduction which is pleasing to the eye, and that's what Phocus does very successfully. Some demosaicers worsen the noise and make it look more blotchy, so the best technique is probably to have a demosaicer focused on detail in the bright areas and one focused less on detail and more on smoothness in the noiser shadows, I suspect Phocus has such a technique rather than using the same demosiacer over the whole image.
Concerning demosaicers specially made for AA-filter-less cameras I don't know if there needs to be that much of a difference, I think Lightroom makes a quite good work in the bright areas, and all have aliasing problems. Maybe Phocus has a little bit less, have looked more at texture so far. Possibly texture is a bit overdone in LR, and that might be that a demosaicer designed for AA-filtered cameras is a bit more sensitive to catch texture, but that's just a speculation.
I haven't had time to make comparisons with my Aptus yet so I don't know what I will think of the absolute image quality when comparing side by side. I hope to be able to do that quite soon, but this calibration data problem will delay me, seems to be more coding than shooting for me the coming time...