Two cases in point. Yesterday I asked for the route from my house to a hotel in Estes Park, Colorado. Google Earth routed me through Wyoming, and Google Maps did not. Using the route adjustment feature on Google Maps I discovered that it reported the Wyoming route as a bit shorter and, since both routes are mostly freeway routes, faster.
Today, I ran the route from the Schaumburg Sports Center -- home of the National Karate Tournaments -- and the Westin in Lombard -- the new home of Windycon -- to see how easy it will be to get between the two next year if we go to Windycon. In this case I started with Google Earth since I don't actually have the address for the Sports Center so it is easier to set the start place in that tool. It routed me down Irving Park road to Medina and then to Lake Street before catching I-355 and Butterfield to the hotel. I then did something that popped up the map in Google Maps, which came up with the route heading up Wright road to the Elgin-O'Hare expressway then to I-290 and I-355. The latter is a more sensible route, regardless of which is longer or appears faster.
My conclusion is that the two tools, while sharing a common corporate creator and some other interaction, have either a different route model or routing algorithm. Since I'm fairly sure that most routing programs still pretty much use a least cost routing algorithm on a weighted graph, I suspect that it is the route model -- i.e. the underlying graph -- that differs.