Yet, you don't believe that if the 28.9-grain load also had an ES of 8 and an SD of 4 that it would necessarily become bigger. More importantly, if the ES and SD numbers for the two groups were flipped, you wouldn't be jumping up and down and saying that tight ES was the key to accuracy, because, as your actual numbers show, velocity is not the dominant term influencing precision here. But trying to see these things is how confusion in our efforts arises. We have to isolate variables from one another to see their effects independently.
Incidentally, I accidentally copied the windage number for gusting 5-7 from 90° at 100 yards when I meant to describe drop. That windage effect over a 2 mph difference is what is the tiny moa. If the gun barrel were perfectly rigid, you would get a drop of over a 100 fps difference of about .17 moa. That's more like what I've calculated in the past. I patched the posts to reflect the correction. But these two numbers are further evidence you are seeing the effects of muzzle vibration in response to recoil moments and pressure distortion.
A really common thing to see is a vibration that switches between vertical and horizontal ellipses as the barrel time shifts, having minimum area (smallest CEP) in the transition between these two extremes, where it is circular. So I looked at your target and saw it seemed predominantly wide at the low charges, gaining some vertical, but also some width in the middle (a bad precision node to my way of thinking) and then back to predominantly horizontal (wide) again at the end. What I look for in that sort of situation is where the two trends crossover, as that is likely to be an area minimus for both modes.
In the plot below, you can see the vertical has a couple of apparent flat spots. One between 26.4 and 26.8 grains and the other at 28.0 and 28.2 grains. In both instances, the horizontal dispersion range crosses them over at about 26.6 grains and 28.1 grains. So I would try those two loads.