My seat of the pants analysis is that a/c efficiency goes up a little with speed, due to the extra airflow over the condenser, but lets ignore that for now. The heat load on the cabin should be constant with speed, so the hp required to cool to a given target temperature is constant as speed varies, and if the bsfc is constant (an assumption but not bad over a narrow range) then the fuel burned per unit of time for a/c is constant. If the fuel burned per unit of time is constant and you increase the speed you decrease the fuel burned per mile in proportion to the change in speed, so the a/c's effect on miles per gallon drops with speed. The extra efficiency we ignored earlier helps make this effect a little bigger since the hp isn't constant but drops a little with increasing speed. One problem is that I did lie when I said that the heat load would be constant. Solar heat load, yes. However, the cabin is cooler than the ambient so there is conductive heat flow into the cabin as well. The R value of the cabin is constant (ignoring the minor nit of air leaks increasing with speed), but increasing speed will transport more heat to the outer skin of the cabin so the conductive heat load will actually rise with speed. Think of the car sitting still, so there is a nice, thick, almost stationary layer of air on the exterior. The temperature of this layer will be slightly cooler than the ambient due to conduction into the cool interior. Now start moving and you blow away more and more of this layer, replacing it with ambient temperature air. How much does this matter compared to everything else? No real quantitative clue, sorry
. My guess is not that much for high R values but lots of the cabin wall is single layer glass with an R value of about 1-1.2, so the external boundary condition will matter more, but it still won't trump the primary effect described above. Anyway, my $0.02 worth of analysis.