TLDR: you cannot accurately determine differences with casual testing like this.
I have experience in vehicle fuel economy testing of aerodynamic devices, and I have high doubts about the testing method as shown in the video.
I have reasons to believe the dimples will not work, will add more on that at a later time.
Especially for relatively subtle changes like this (compared to drastic vehicle shape changes), a precise method is needed. With a few short runs like this, the uncertainty in the result is just too big.
When we did fuel economy testing, we would drive in highly homogeneous conditions (i.e. both test and control vehicles at the same time on an empty oval track). Still, there are things you cannot control such as wind and weather.
This means that we would have to drive for hours on end to collect enough data to even out the conditions we cannot control. Even then, we would routinely discard data due to non-constant conditions.
They do not mention the distance or the time traveled. However, we can estimate it.
With a total fuel usage of 478 grams, and a fuel density of 748.9 kg/m³ (source), it would amount to 0.638 liters consumed. Assuming that is for the fuel-efficient vehicle (which consumes 7.93 l/100km), that would amount to 8.04 km driven, which at 65 mph, would take 4.59 minutes. They mention a sequence of 5 runs, meaning that they probably slowed down and turned around 5 times, which further reduces the stability of the measurement. This is such a short measurement, that a small change in conditions can seriously influence the results.
Also, ICE engines take time to stabilize, so we would run the vehicles for a significant time to warm them up to operating conditions, and let the fuel usage stabilize before we started the measurements.
Doing all this allowed us to reach reasonable accuracies (of around 1-2%). If they did not do any of this, their uncertainties will be much much bigger.
Furthermore, if you use two different vehicles, even when they are the same type, there will be significant differences in fuel consumption due to manufacturing tolerances, different wear and tear (of for example tires) etc. So it could very well be that the dimpled car was more efficient to begin with.
For the interested reader, we've described our testing procedure (including correction for differences between the vehicles) in more detail here, specifically section 5. Track Testing