It looks more obvious to me and less confusing. I guess it can save some time as well, having two paths to navigate on the tabs panel.
So, collecting some feedback (the comment from bad π, my own opinion, and the comment from cwallenpoole), this could be the right way of moving with the down key:
The last arrow (from the last cell to the first one) is optional here (now it blocks when at the end).
The up-arrow would go in the inverse direction, of course.
I'm not sure if it is sensible to also map the left-right arrows to horizontal movement - I think, now they are moving in the text field.
Two-dimensional navigation here would be complex and confusing. I think Paulo's screenshot is a perfect example of why.
We also want to emphasize the fact that the list is ordered; if you pressed ↓ and it went to item #4 on the list instead of #2, that makes no sense.
Thus, like the mouse wheel on your mouse, we lock to 1 axis. Press → or ↓ and you get the same result -- moving to the next item in the list, in order.
Pressing down and seeing it move right is definitely not the expected response. Pressing down (and up) as a means to select from an auto-complete dialog is expected.
When pressing the down arrow could the selection move down and upon reaching the bottom return to the top of the next column (top to bottom, left to right).
I'll agree that on its own, especially since there are only at most 6 cells, it does have a more intuitive feel to have up move up and not left, and down move down and not right in this layout. This feels especially pronounced when using the edit tags functionality.
But, at the same time, in Google, when I want to move between auto-completes, the down arrow is how I will move between selections. Because I'm thinking of it as an auto-complete (and not a grid layout) exclusive use of the down arrow "feels" more intuitive in that context. When creating a question, the "it feels like auto-complete", to me, at least, becomes more prominent.
Also, despite its layout, ideologically, this is really a
Personally, if I were king of the SE, I'd stet. It does not cause substantial cognitive dissonance when thought about, and it follows (at least in one sense) a very common UI already in place.
(This originally started a comment, but it seems important enough to append to the answer)
This actually goes to the question of ease of use vs. intuition. The easiest motion for a person is to continually hit the same button and possibly another. The more expected UI is to require the person to (probably) use 2 buttons and then make them need up to 4 to get full motion.
I prefer one button over two because I find it easier and, honestly, I like my UI as easy as possible even if that sacrifices a more intuitive interface. I'll only find an interface strange perhaps once or twice and then I'll adapt. I'll find a less easy interface less easy every single time I use it.