Flats issues - imaging train sagging?

Recently, I took for the first time LRGB images with the super reducer. And I noticed severe vignetting in the corners:

I guess, I should have used square filters and not the round ones… But anyway, I should be able to remove this vignetting with flats:

Calibrating my images with these flats seemed to work. But when I stacked these subs, I images like these:

In some corners, calibration worked well, but in others the flats clearly failed making the image either too bright or too dark.

I checked  the calibration, took new flats, took panel flats - but the result was always the same. I wonder if my imaging train is sagging and that causes the incorrect calibration.

Asked on the ccd-newastro mailing list. Ron Wodawski was incredibly helpful:

• he figured that the reducer should be able to illuminate the entire chip easily (it is specified as delivering a flat field of 65mm diameter - more then enough for my 16070 chip)
• he said to focus on all elements (correct orientation, correct spacers, correct location)
• he also said it could be because of the dew shield - that's easy to test.