Thanks for all the feedback to this thread. I've done lots of test and reading etc the last couple of weeks, and here's my resulting Argyll recipe and related info, I hope it will be useful to others starting out with printing:
http://www.ludd.ltu.se/~torger/photography/argyll-print.htmlThe summary workflow is:
targen -v -d2 -c <preconditioning.icc> -G -g32 -f840 <name>
printtarg -v -iCM -h -R1 -T300 -p A4 <name>
chartread -v -H -T0.4 <name>
colprof -v -qh -r1.0 -S AdobeRGB.icc -cmt -dpp -D"<description>" <name>
or colprof -v -qh -r1.0 -D"<description>" <name> if you can live with that perceptual/saturation mode works as relative colorimetric with BPC
profcheck -k <name.ti3> <name.icc>
I found that preconditioning is good to use, and you don't really need to make your own preliminary profile as targen only needs a rough idea of the properties, so you can use the maker's icc profile, or even just another profile for a similar paper. What preconditioning does is that the patch distribution is more perceptually optimized, which means you get higher density close to neutrals, and a bit lower density in saturated greens etc, which I think is a good balance. I've added the usual array of gray patches (but not too much overkill like I've seen in many other recipes) to get some additional precision for monochrome prints.
Regarding patch count I chose 840, it makes even 4 A4 sheets with the Colormunki. I made comparisons in the range 210 to 1680 patches and look at other users experiences with this printer type and how many patches professional software has. My conclusion was that 840 is "as good as it gets" without overkill, if you like to have a wider safety margin you would double it to 1680, but it's really tedious to scan with a manual instrument like the Colormunki and the return is tiny.
Then I've added the profcheck step with DE2000 to sanity check the profile. You do get DE numbers directly from colprof, but I prefer to list them in the more perceptually adapted CIE DE2000 unit.
I've increased the smoothing from the default 0.5% to 1.0% in colprof. During testing I found that very little accuracy is lost and in some occasions you get a little bit smoother profile (smoother gradients), and as my primary use case is fine art printing I can lose 0.2 DE if I gain smoothness.
I've made some experiments with perceptual vs relative colorimetric. The original ICC intention says that relative colorimetric should clip hard at gamut boundaries both towards saturated colors and towards shadows, and if you use Argyll's cctiff tool it does exactly that. However, typical printing software like Lightroom and Photoshop will roll off slightly at gamut clipping and will always have BPC enabled which make relative colorimetric similar to a perceptual mode, especially if you have a semi gloss or glossy paper where the gamut is large enough to not clip much. The problem with perceptual mode is that it's entirerly up to the icc profile which needs to make an assumption of the source gamut (AdobeRGB in the above recipe) and it does not always match so it's a bit less well-defined than relative colorimetric. In practice I think there's little place for the perceptual mode and I prefer relative colorimetric. If you generate an ICC profile without a specific perceptual table your profile generation will go much faster.
Concerning OBAs I have not gone too deep with that, but if you have a Colormunki instrument you have UV cut and it's not much you can do about it. It seems to me though that in practice compensating for OBA becomes really messy as it requires a well defined static viewing condition, so I think it's better to use low OBA or OBA-free papers and not care.