For someone who was just telling us that statistics are 88% made up, you seem suddenly quite happy to accept them uncritically when they support your case ;-)
I have a few questions, because frankly I doubt it's nearly as simple as this graph makes out. Apart from rolling resistance and aero drag, the other things that are going to affect efficiency are (a) constant factors (e.g. pumps and whatnot that run all the time the engine is on); (b) the most efficient speeds for the engine. That being so, why are there only two peaks on this graph and not five?
The indicated source of the data is the EPA. According to teh interwebs, the EPA tests are conducted on a rolling road (1) at a constant 56mph; (2) using a simulated "urban cycle" which is stop-start driving between 0 and 30. In other words, they're probably not even
testing driving at constant speeds under 30mph using the appropriate gear for driving at that speed, all their data points at low speed are measuring consumption under acceleration. Which is going to be much less fuel-efficient. Incidentally if you built a similar graph from "crowd-sourced" vehicle telemetry data you'd probably still get the same problem, because who drives at a steady 20mph for any length of time?
http://www.fueleconomy.gov/feg/fe_test_schedules.shtml
I'd also be interested to know how much the constant factors actually account for: if you put headlights and AC and radio on, low speed driving is going to look worse. Things like the fuel pump, these days, ought to be PWM and therefore vary with engine speed. Cooling fans - 10A when it's on? 120W, so about the same as headlights...