Thank you all for helping me challenge my thinking and workflow. Too easy sometimes to just plod on with something that works, but may not be the best approach.
OK, refining things a bit:
here are the tiffs from which I created the jpegs above. Should have done this in the first place. Except for the linear source file they are all in ProStarRGB, which is another thing I'm playing with at the moment. In this situation it's just a container - I've performed no adjustments after conversion from the scanner profile so it should not be influencing these results.
The LR downsample+jpeg+AdobeRGB process contributed to the clipping and obscured the actual results I got. I also capture sharpened the files, 0.6 87 Smart sharpen, which probably added a little clipping too as I did not protect the highlights. lazy.
Yes I do have clipping at the upper limits. No it's not clipping that I'm worried about as the files print fine. It's mostly in the paper colour background, touching on the brighter colours in the drawing. The approach I used in the end was what similar I described a few posts above: CoCa DCSG profile extended by setting the whitepoint at 1.1. This gave me a slightly compressed masterfile that I then brought back to the exposure I wanted, while trying to avoid clipping. It's a controlled approach: I can apply as much compression/extension as is necessary to capture a given original. The Silverfast output is definitely safe, but will create unnecessary correction work if the gamut of an original fits well within that of my target.
A gap in my understanding: I'm not sure what you mean, Mark, by 'the gamma factor used in my profiling and scans'. I'm not applying any gamma adjustments to my scans. I'm relying on the profile to restore correct exposure to the linear source file. I output 16-bit linear, assign the profile, convert to working space and the exposure and colour are 95% of the way there. The Silverfast and Epson scans have been gamma adjusted, but out of my control.
I tried what Dean suggested: use LR to bring the HCT Extended sample close to the look of the 'final colour' sample. It came up nicely, retaining good variation in the oranges that is a little lacking the CoCa-based final. I suppose the measure of a good profile is not just how close it gets you when you assign it, but how well the resultant file responds to adjustments. How robust it is. HCT-based profiles yield very robust files, for reflective and film work.