If you look at it logically, the shorter span direction is the stiffer and the same thickness slab in that direction is significantly stiffer than in the longer span direction, so it will tend towards the logical "support direction" with the long span being the "one way" direction. As the short span becomes shorter compared to the long span, the slab tends more and more towards one way action in the long span direction and the short span becomes more and more like a "beam" support.
The longer span direction should have least cover so if the bands are in the short span direction, then the distributed tendons would be above at the supports. If the banded tendons are in the long span direction, they should weave to be at minimum cover at the supports.
So technically, the shorter direction is the banded direction. Interestingly PTI logic is (or used to be) the reverse and it is difficult to understand why!
Even in band beam and slab systems, the band beam in the short direction is still the logical option, even though most designers have not figured this out yet and treat it like a beam and put it in the long span direction. The band beam is a slab thickening, not a beam, and the band in the short direction is more technically correct acting like a drop panel, and no more expensive. It is not until the beam is actually stiff enough to act as a real support for the slab in the other direction that it should be in the longer span direction.
Not that I would recommend anyone ever using the banded distributed logic in a flat slab, being a bonded PT person who along with Hokie prefers column/middle strip logic in both directions. If that method is used, service stresses should be kept very low (much lower than ACI suggests) to avoid unwanted cracking and extra deflection.