I've been doing something very similar for years. Not with an iSis. And for doing the comparison it's worthwhile noting that it should be calculated with a dE formula. dE(76) for proofing simply whether the printer reproduces the reference values correctly, unrelated to impacts of human visual perception, is fine. dE(2000) is a far more complex construct that corrects for the phenomenon that dE(76) fails to take certain distortions of how humans perceive colour relationships into account. I have also determined in my article on Extended Printer Accuracy Testing (this website) that while the 24 patch CC remains a useful check on accuracy, one learns more about the robustness of this accuracy across the gamut of the printer/paper combination by constructing and testing more challenging targets that remain within the gamut of the output device.
Interesting article, Mark.
I like the notion of making patches specifically for the lighter pastels as was as the near gamut, highly saturated colors. Very cool.
I've explored an expanded gamut test by making a Lab grid every 10 units, like Lab(40,-30, -50) over the printer's gamut. Also, like your work, it's printer/paper specific and is derived from the ICC profile. It's fairly automated using Matlab to create a full Lab grid then removing the points that are out of gamut leaving behind the in gamut grid points. Then the patches, around 400-500 or so are printed, scanned, and stats are calculated. However, it includes fairly evenly spaced colors within the gamut and isn't focused on the lighter pastels and saturated colors.
I like your specific approach of targeting these colors. Perhaps I'll look at modifying my grid approach to focus on these colors. Particularly the lighter pastels that are near the L* gamut. I've not focused on this but believe it's quite a good area to look at.
As for dE 1976 v dE00, I do find the dE00 more useful because it is a good perceptual metric. Near neutral colors tend to have larger dE00 values than dE76 while even moderately saturated colors have smaller dE00s. My 9800 has anomalies in some areas of neutral colors and dE00 is somewhat better for these. In particular because at both low L* and high L* dE00 is less sensitive to L* errors but much more sensitive to a* and b* deviations which is a bigger problem for the 9800.