Are you talking about the visual information or the game information or some combination of both?
I know that some games keep two maps, one that contains the visual information and one that contains the game information. Using the terms you gave above, the game information map would be scaled per square while the visual map would be scaled per tile.
Have you considered the z-order issues and the possibility layering multiple tiles? Say I'm building an RPG and I drop a tree on the map. I want the leaves and branches to be drawn on top of a character walking beside the tree but the ground still has to appear below the character. What about the trunk of the tree? In an isometric engine it may go behind or on top of the character depending on where the character is.
You can simply that sort of problem by saying the tree has to be game object and that terrain is a single layer drawn at the bottom. But going that way can cause many games to be forced to create a large number of game objects that are really static terrain features.
Jay