I was working on learning how to tackle "stepped" building shapes when I ran across the difference in the make-up of building shapes created from a CGA script (or one of the "Building-from-Footprint" scripts and the buildings that I created manually. The script generated buildings seem to be created from the "bottom-up" while the manually created shapes are "top down" I've included an example of the two methods side-by-side.In the "Building-from-Footprint" method it's like I'm building the building shapes upside down. This has all kinds of ramifications.
It sounds like....
The CGA creates a box from 6 squares, each with 4 vertex.
The shape creation tool makes a proper box that is a coherent box.
On export (collada - i haven't looked at other formats) there are options for "vertex precision" and "merge vertices within precision" - I guess that playing with these can help reduce vertex count.