Sorry, maybe you didn't see that I already tried tried that as stated in my post...
Sorry, I missed that. It would help a lot if we could see the image and get a little more info on what you're trying to achieve.
I just played a little to see what I might be able to come up with, and one thing I noticed is that the green channel seems to be the very worst when it comes to banding issues. When amounts of green are mixed in with the other two channels in a gradient, it "pollutes" the area and bands occur almost every time. When I made pure gradients of 255 blue to 1 blue (effectively black), and the red and green at 0, banding couldn't be seen, unless I was only looking at the blue channel, just to see what was there. In RGB mode, all was well. Same when I did it with red at 255 and blue and green both at 0. Even with red and blue both at 255, as long as green was at 0, things looked pretty good.
I found that interesting.