When you do the gradient test, you assign display profile to the gradient, so the monitor is L* calibrated, and the image is in L* based color space. But when you're working with AdobeRGB images, they are rendered to the gamma 2.2 based color space. To simulate the result you may assign AdobeRGB to the gradient, and check if it's still so smooth
That's not the way the test was designed to work as I understand it. A normal image in a normal RGB working space is always translated (behind the scenes) to the monitor space so you can see it. Hence will always suffer from some degree of banding or other gamut compression byproducts. The test gradient is not a normal image, it is designed to bypass Photoshop's translation from an RGB working space into the monitor space. IOW, a representation of a monitor at its absolute best.
When I run a monitor cal/profile at 2.2, and do the same test, assigning that 2.2 profile to the test image, it still looks very nice, but there is a bit of a hiccup/band in the 1/8 tone area that goes away when the monitor is operated in (and test image assigned to the ) L* profile. Just seems like L* makes the monitor run a bit smoother IMO.
When I get back to the studio tomorrow, I'll leave the monitor in its L* profile, do the gradient test, leave it in my AdobeRGB working space and take a close look.
What would be the recommendation for calibration when the working space is ProPhotoRGB which is Gamma 1.8 ?
In color space terms, the gamma has to do with how edits get applied to the file with respect to shadows/mids/highlights, so that's not the same application of gamma based on what I read here:
http://www.adobe.com/digitalimag/pdfs/phscs2ip_colspace.pdf , pg 5.
(I'm sure Andrew will jump in here, he wrote the article!)