I've seen cases where the soft proof shows some banding but not on output so this is very odd.
Do you know anyone nearby with a ColorMunki who could measure the targets to confirm that the issue is with your unit?
Have you tried using the iteration process whereby you can create additional targets (more patches) after building the first profile? Its possible the 100 patches are insufficient to produce a good profile. I've yet to see this, but I can't say I've tried more than a few papers with the Munki. Try to sample some of the problematic colors and use them such the software will build a new target to add to the existing profile.
Unfortunately I don't know anybody else anywhere near me with one. Honestly I don't think that we're in iteration territory here. Also, the fact that I got a good profile from the Innova F-Type some months ago, but now I can't repeat it, tends to indicate something is wrong now that was ok before...
Looking at the gamut plot clearly shows the issue, so why it is not showing up in soft proofing is just weird.
If I get time, I'll try print off a couple of examples using the Quato test target, scan them, and post them. Obviously a lot will get lost in translation, but the scale of the shift should be evident.
Good thing I'm not doing this for a living :-)
I've raised a ticket with Colour Confidence, the reseller, so maybe they'll come up with something.