Yeh, I had considered this as a special case of irregular, though obviously that's not strictly accurate. The reason I was thinking of it that way was pathing-centric, since you'd want to consider macro-tile (in some sense) sized regions of micro-tiles that were essentially homogenous, as one macro tile for pathing calculations (for scalability reasons), which means that (from a pathing perspective) you wind up with what amounts to an irregular grid (in some areas you just treat it as a macro-scale tiling, in others you need to dive into the details). Also for display purposes it's pretty much an irregular grid, because the snap-to scale would probably need to be more at the stack-size level than the micro-tile level or else movement points would seem a bit un-natural for the player.
Anyway, I'm leaning towards regular grid with overlay layers, because it seems adequately flexible without MUCH added complexity, and because it lends itself to staged implementation (regular grid is all you get in version 1). On how to tessellate, I'd prefer hexagons since neither is rally any more work than the other and I think the end result is, on balance, superior.