I try to understand the interaction of (real) engine + gearbox on the consumption. So I purposely don't want to account for friction on air and floor for now.
For the engine alone, the optimal efficiency is obtained for some given rotation speed ( about 2000 RPM, I guess ), then the consumption per turn increases above as well as below this optimal.
Now if we couple a gear box, as the wheels velocity slowly increase (slowly for quasi-stationarity) we have to change gear. During this slow process the engine thus cycles between under-rev, optimal, over-rev several times.
→ at the end, might it be that the consumption per km/h varies non monotonically with the wheels velocity ? (again, not accounting for air and floor friction. Could be tested on rolls.)