Every data analysis done in good faith has strengths and weaknesses. As you point out, a mileage normalized result would appear skewed to an average driver if a high mileage extreme driver was in the mix. Since I think that scenario less likely than a result skewed by a bunch of drivers who record data for a couple of tanks in one season and then stop, the proposed analysis has merit.

And in any case, I am not suggesting one or the other, but multiple views that help point out mileage performance obscured by our as yet relatively immature data. Incidentally, since I think the data will turn out to follow a normal (gaussian) distribution, over time the mode and median will converge, as will a result that has been calculated by mileage normalization.

Your comments regarding trip time and temperature are spot on. However, if you have 5 variables, taking 2 into account is better than 1 .. 3 is better than 2 ... etc, assuming equal weights.