Hello Paul,
There are two things:
- the not perfect RGB/Gray conversion done by ACE for the above explained reasons, affecting the very deep shadows. There is no easy way to avoid this, but the visual real effect could be considered negligible for a single conversion in my opinion.
This is somewhat confirmed by the fact that your crops are looking nearly the same before soft proofing and you can spot the slight difference only by looking at the histograms.
So in any case I think this aspect is not something to be really worried about.
- the noticed difference in soft-proofing (and real prints to some degreee, as you have found) in the shadows zone between the same RGB and Gray converted images, even with colorspaces matching the same gamma value.
In addition the results are varying depending on the ICC profile used.
This is very tricky, and the fact that this is not only a visual effect but it is somewhat reflected in the real print is not so easy to explain, given that the source files are looking nearly identical on screen before soft proofing.
I hope someone more skilled than me cold put some light over this aspect.
In the meantime the further only thing I can suggest you is to try to print the RGB/Gray test crops even using "Perceptual" intent and NO black point compensation, and see if the printed results could be better matching the original file as seen on screen before soft proofing.
Maybe this workaround could mitigate the difference to an acceptable degree, let we know.
As last resort, if you could consider feasible avoiding the possibility to have split toned images and to soft proof the preview, try to bypass the ACE at all and print the RGB/Gray test crops letting the printer to manage colors and engaging ABW (neutral dark setting to start) on the Epson (I don't know if there is something equivalent on the big Canon). This will give you a bit of higher dmax too. it could be interesting to see if the printed crops in this case are better matching the files on screen. If the case you have at least another workaround available, I know that this is not ideal but it's still better than none.
Any further comment/opinion is welcome.
Ciao.
Andrea
EDIT: I'm suggeting to use "Perceptual" without BPC and not "Relative colorimetric" with BPC,
as wrongly written previously and as now corrected. Sorry for the typo.