The issue will not be resolved, but it will be less severe. That is hard with a circle and using n images. They are mutually overlapping, unlike a linear set of images. I am not trying to make a helix. I worry the handedness will be apparent.

This is the simplest case. You might imagine a non-radial example. There, it is easier to do overlaps of transparency.

Also, then you would have to manually size every image instead of them all being the same input size.

I would prefer Adobe just fix the bug. It is just math for them.

I find it non-intuitive than an opacity mask would be in CMYK color space, either way. One does not color
manage an opacity, which I presume is the problem. I suspect they did it that way to avoid handling the creation of a mask without crashing or converting anything.

This would be my approach. Instead of using overlaps, I keep gradients within masks where needed only. Semitransparent on opaque gives better results.

That was my initial thought as well, but the results are pretty good. See the attached for how it blends at the center. You can tailor the blending effect there by adjusting the corner radius. The image choice radically impacts the visibility as well as the background color because of how they bleed through.

Dark images in the lower layers can be paired with a dark background color or even the same image with no mask. See the bottom two images, which vary by the background color (dark blue vs white). Effects vary because of the lack of uniform gradient behavior. Where things become challenging is in blending images with light and dark backgrounds. If it is not done correctly, the perceptual weighting of an individual image becomes high, and attention is drawn to one image, instead of being balanced.

There is a small Gaussian blur to eliminate the banding in the line gradient. That is not the issue. The same things happens without it.

The bottom circular example is the general solution, provided that the gradient opacity issue is resolved.

I can imagine what it looks like, yes. But with images the effect of the weak middle where gradient overlap would be much... weaker, right? Can you share a fragment of the collage, just to see how the problem affects the design?

I am basically creating blurred transitions between radially arranged images that mutually interact inside complex shapes (not a circle) to make a kind of subtler collage. The number of images typically range from 3-6. Photoshop can be used, for simpler shapes, but for complex geometries, Illustrator is much easier. As you alluded to, the challenge is in visualizing the hierarchy of the images and masks if one does not do this routinely. This is much easier to see inside Photoshop. I can tailor the background color to work since the images are branded that it is less obvious, but it would be nice if the opacity masks did not require that for a more general solution. Once I have the masking set up, I can build an entire masked image set within a few minutes.

Yep, I know about this myself too.
Even with CPU Preview (GPU speed performance may alter colors a lot), both a screenshot of the canvas and an exported TIFF (and PNG), measured in Photoshop, give me G and B climbing up in the middle gradually in sync, up to 64.
I’d agree 50 + 50 should give 100, but... Math... Gamma... Color profiles... too many factors contribute.

I wonder what are the things you do often that make it important? Just curious.

The parent report:
Gradient across stroke, inside opacity mask is thresholded for half the width of the stroke
http://illustrator.uservoice.com/forums/601447/suggestions/51190204