I'm glad you ran some tests yourself, however...
Two to three s-units difference? I think there was (is) something up with the s-meters in question. I don't doubt that you got those results, however, I don't know that I would trust them at all.
To make 1 s-unit difference it takes 4 times, 2 s-units takes 16 times, and 3 s-units takes 64 times the effective power difference.
I have seen no believable data anywhere ever that would lead me to believe for a moment that adding a second antenna would have anywhere near this much of a gain, especially means that you were pointing the antennas exactally opposite of the direction you should have had any gain at all.
I'm also concerned with the lack of consistency of your results at different distances apart. For example, at your 270 degrees, you moved further away and the received s-units are higher then when they were closer.
Also, while I am not specifically sure with the radios in question offhand, when most radios get to 9 s-units the scales tend to lump bunches of s-units together, and as such tends to get much less accurate, not that most of them even resemble being accurate to begin with.
I though I mentioned about s meters early on in this thread, however when I went to look for it I now cannot find it. I would never take anything a s-meter shows as written in stone. In most radios they are wildly inaccurate, in some cases even after being aligned. Also for testing purposes such as this one they tend to require to much power difference to show any appreciable results.
I have noticed in your pictures the way the antennas are mounted are likely not the best for a single antenna which is also much more dependent on counterpoise than the cophased antennas working together. Mounting on posts such as those you used will definitely have an effect. That being said, That doesn't explain two much less three s-units of difference.
If you were anywhere near me I would offer to try and duplicate these tests, and based on my experiences I also doubt that they would repeat themselves.
I applaud your efforts and am glad you took the time to run your experiment, but it simply doesn't match with what I have seen on countless occasions myself.
The DB