Wolfgang, May I suggest a way to proceed by evaluating a vignetting curve and applying it to your problem panorama project:
If you have a nice blue sky during the day, go to some location with a clear view of a reasonably distant horizon (to avoid parallax effects).
Take 5 or 6 shots (raw, portrait, not bracketed, and no need to use a tripod unless to avoid camera shake) with the same aperture and focal length setting you have been using, full manual mode (no auto settings), and overlapping by 50% approximately. Angle the shots tilted upwards with just enough of the land below the horizon included at the bottom of the frame so that there will be features available for control points when the images are stitched together.
Convert raw images to tiff with the bare minimum of adjustments (no auto settings or vignetting adjustments).
Load the tiff images into PTGui and align with control points only on the distant horizon and optimize vignetting correction on the Exposure/HDR tab (in the automatic exposure and colour adjustment section). Hopefully, a good rendering of the sky will be obtained, in which case save the vignetting settings. If banding problems remain in the stitched output, then there's no point in carrying on to the next steps!
Assuming all has gone well, apply these saved vignetting settings to one exposure set of images from your difficult panorama - i.e. a stitch with no HDR involvement.
Set the blending mode in the Panorama Editor window to "unblended", which will make brightness differences between neighbouring images very sharply evident.
Try optimizing just exposure on the Exposure/HDR tab, or if necessary, tweak the exposure offset parameters of individual images to match the images one by one to eliminate visible image boundaries. If it's possible to achieve this reasonably well, then when the blending mode is switched to blended, the result should be very good.
Based on previous reports, it seems likely that the result will not be perfect either, but if it is, you should be able to deal with all the images in an HDR project.
John