One possible explanation is that the GOM estimate is based on previously achieved mileages, which will be low if they are generated in colder weather. Now that it is warmer, they will be improved, but the GOM algorithm may still retain some bias from the winter memory, which will slowly be overcome as the older values fall out of the stack. Come the winter, you should see a slow reduction of GOM value, as the summer values drop out, and you may actually experience worse performance than the GOM predicts. The effect would be more marked if the change of seasonal temperature is more abrupt.
A possible reason that I haven't noticed it happening is that the seasonal temperature variation here is MUCH less than in Colorado. Well, it's a theory