Ah, okay, I'm understanding this differently now.
What happens is the objects are actually sharing shape nodes AND transform nodes. I didn't see how two objects in two DIFFERENT positions (e.g. translate) could have the SAME transform node, and so have all the same translate values. BUT, each value in the transform node is RELATIVE to the parent, so of course if the transform of the group is different, that will influence the children.

So I was wrong Darkware, thanks for enlightening me Dave. What this means is if you change the position of any of the children, each of the instances will also change position because they share the transform node. Neato.