I've noticed that most unusual routings are corrected (sometimes causing other insanity further downstream, though) when another point is added and re-routing is needed - this happens whether you select a point to add while planning the route, or as I'm prone to do while travelling, using the GPS to reroute from my current point, assuming I'm already on the "non-loopy" route.
Algorithm: Definitely skewed. It might also explain why, when travelling across Canada (Moncton, New Brunswick to Vancouver, BC, for instance) the router (on both Shortest- and Quickest Route preferences) will, annoyingly, divert you into the states at every possibility, even if it adds hundreds of miles. The algorithmic "weighting" of the roads differ from US and Canadian sources.
What S&T denotes as a passable, but unpaved road in the US is shown as being the equivalent of what may be a roughcut logging road in Canada. Experimenting with S&T2007, I followed two routes with my preferences for nice, wide roads turned waaaaaay down; once for a trip in the US, and once in Canada.
What I noted was the roads in the US were generally about a grade higher than the "matching" roads in Canada - but I also found that the best Canadian highways weren't given equal values to their US counterparts - occasionally with reason, but if a route is safe at 110km/h, it ought to be considered roughly the equivalent of one safe at 65mph.
Rant rant rant, gripe gripe gripe. Still love S&T.