There's two things I found surprising. That they tested so many circular arrangements, and that there was a measurable difference between the two 3x3+1 orientations.

If the author of the post is reading this, the image of the two vertical columns is wrong. On the site it's essentially two columns, in the paper it's five rows. That is rows 15 26 37 etc compared to 12 34 56 etc. It makes for a surprisingly large difference, at least to me. The former looks completely unusable, the latter seems quite ok :-) So maybe I should not be surprised that the two 3x3+1 setups didn't perform the same...