warning a dead horse is about to be beaten...

the only reliable results come from a *controlled* test. one where all variables are precisely controlled, not just the driver. lacking one, even small variations lead to fairly large deviations. deviations of 10% or more are not unusual. saying you drove the two cars in exactly the same manner may be true. but, there are other factors that will affect your results: the route, light timing, road conditions, traffic, temperature, humidity, tire pressure, engine wear, electrical load, oil type, oil condition, type of pavement, head winds, cross winds, etc. all need to be controlled and the same. if they're not, comparative results are virtually meaningless. take head or cross wind for example, even a slight cross wind during a portion of the "test" will have a noticeable affect on comparative mpg. cross wind is not something you or anyone else has control over when driving the car outdoors. if you want to see just how much cross wind alone can affect mpg, there's an excellent prius calculator that shows you at:

http://privatenrg.com/#WindBigDeal

from this report, you can see that an imperceptible 8mph cross-wind can affect the mpg in a simulated (read: controlled) test by more than 10mpg. and that's just one variable. there are dozens of such variables and most of them you have no control over.

my point is: lacking a controlled test, the standard deviation for aggregate data (aka, this database and others) tends to be minimized because variations in all the variables tend to cancel themselves out across comparative samples. therefore avg. results are more reliable. that's a simple statistical truth.

okay, the horse is now dead.