Heightmap based 3d would be very simple to add via something like f(x, y) -> x, y, heightmap(x, y). The city building genre is essentially "2.5D", as the gameplay itself is purely 2d with the third dimension existing solely in the presentation layer.

Edit: As for the roads on uneven terrain, I'm thinking about deriving z coordinate from another 2d Bezier curve traced along 't' of the road. Control points could be derived from a convex polygon obtained via the road/heightmap intersection.