Looks like a three way tie, so you'll have to make at least one piece of each! I **strongly** dislike stackables for reasons I've already outlined. Completely modular ala DL205 is my favorite, but for small projects I certainly don't mind the expando-brick like the 06 or the early SLC's.
I like the idea of the a number of different slot-count bases, like the 205/305/405, etc. The only advantage (other than potentially cost) to the stackables is finer granularity of I/O count vs. cubic and square inches of panel space, and if you have 4/6/8 slot bases, at worst only 1 empty slot is forced upon them except in extremely small jobs. And don't forget, space for future expansion is generally considered an advantage, not a disadvantage.
I once had a guy from Grayhill trying to sell me on why it was good to use 1 or 2 point modules because you could build the machine with the exact I/O count. (Even though their cost/point was enough higher that all the money saved went to Grayhill, not to me, and at the cost of shafting my customer. They never seemed to notice that point.) Then they came out with 16-point modules. I asked (already knowing what answer I would get) why they thought 16-point modules would be a good idea. He said that the the cost/point was enough lower that I could save money overall, even if it meant I had to ship machines with unused I/O points. I asked him if this meant his competitors had had a superior product all along.
He said that no, low density modules were still superior to your high density ones, or AB's, due to their low point count, but that HIS multipoint ones had advantages over his low density ones, because of their higher point count. I went for a shovel.