Jonathan - I much appreciate your responding to this issue in the tightly reasoned manner you have; it is an intriguing problem and you have raised additional considerations about it that deserve to be probed. Notwithstanding the advice that I reported here - which under the circumstances I still think is plausible, in the back of my mind I haven't dismissed earlier suspicions that perhaps there may be some colour management issues relating either to the scanner software, the scanner itself or the combination. So let us now drill down a bit more.
It gets a bit difficult to analyse this without knowing more about how the scanner software ACTUALLY works, internally. But the software on the disection table here is Silverfast Ai 6 Studio Edition. This is supposed to be the cat's meow of scanning software (unless you own an Imacon), but sticking with the animal analogy, it is also a dog's breakfast for understanding what it really does, because the documentation is crummy and the interface, to put it politely, is "arcane".
Let us recall that I am scanning colour negative material, and that has limitations for colour management different from positive films. It has an obtuse colour management options page that lets you make the following selections:
(A) Colour Management: (A-1) Internal>Monitor: either None or ICM (Image Colour Matching); (A-2) Internal>Output: RGB, or ICM, or CIE Lab, or CMYK. For A-2 I use RGB. For A-1, I can use None or ICM. ICM yields a much more saturated image on screen. But interestingly, this choice does not affect the RGB colour values. (By the way, I haven't seen, or perhaps missed, where they define what "Internal" means.)
( Profiles for ICM: (B-1) Internal: The choices are a hodge-podge of colour working spaces, such as RGB98, ProPhoto, Colour Match RGB, etc. I have used either RGB98 or ProPhoto as we discussed earlier, and this is the choice that critically affects values once the image is opened in Photoshop. (B-2) Rendering Intent: it provides the usual suspects; I use Relative Colorimetric.
© Embedded ICC Profiles: Once you select a colour space under B-1 above, this section automatically repeats that selection, but there is a check box to "Embed the ICC profile", which I leave checked because that is what codes the colour space data going into Photoshop.
Now, for dealing with the special case of colour negative film, Silverfast has a module called "Negafix". Negafix allows you to select from a very large number of colour negative films according to what is exactly the film you are using, or comes closest in terms of hue/brightness/saturation on the monitor (mine is very well-calibrated and profiled). Then it allows you to tweak the generic exposure and hue associated with that selection, just in case - like me for example - you are using a brand that doesn't come out bang-on correct for any of their presets. Not explained is how this Negafix module integrates into the scanning workflow - i.e. does it act like a film profile at the outset of the scanning process, or is it converting numbers at the end of the process of generating the image file from scanned data (I guess what you mean by "cluster-fudging").
This is as best a description of the Silverfast colour management system for negatives that I can give you, after having read the Silverfast manual and Taz Tally's SIlverfast book, and Ian Lyon's Silverfast tutorials and watched Lasersoft's crummy Quicktime movies that are supposed to "explain" things. I frankly think there is confusion of concepts in the layout of their whole colour management set-up, and I have told them as much but there has been no reply to that generic complaint. (Be all that as it may, I must say that save for this issue - and one other immediately below - once you mess around enough with the settings and get a combo that works, it works pretty darn well.)
Now Jonathan, the issue of "red" was stuck in my mind for the other matter that I sent you the image snippets about - and I asked myself the very question you are now posing - what if the colours are indeed out of gamut, or put otherwise what if the ProPhoto selection is allowing-in out of gamut colours. So after seeing how much better the skin tones emerged due to a simple switch from ProPhoto to RGB98, I went back to that red/purple issue and rescanned one of those Chinatown red images in RGB98; lo and behold, the red came out tamer and when soft-proofed, there was still alot of toning down with some apparent hue shift toward purplish, but not nearly as bad, and not to the extent of wiping out image detail that happened when switching on soft-proofing with the ProPhoto version. So I think this indicates that there may be a gamut issue at play here affecting both problems. In fact, it may be one and the same problem. It seems that selecting a narrower colour space when dealing with some troublesome colours squeezes the latter into gamut more successfully than can happen once the image is opened in Photoshop with the wider colour space. I'm aware that I'm treading in a bit of an alligator swamp because I don't know all these alligators and their habits as well as would be desirable; so now that you've read in much more glowing detail what I can tell you about the matters you raised, I am very interested in your assessment.