When I do testing, I do it at the speed at which I'll be driving: 55mph (plus or minus a bit.)
In order to extrapolate the data to different speeds, you really need multiple data points. Some engine mods or general driving habits may scale linearly with speed... aero mods would probably not be so linear. For each change, a minimum of 3 data points (3 different speeds) would at least point to the type of scale to expect. Actual results will vary significantly from driver to driver and car to car (and with a whole slew of environmental variables.)
I guess it comes down to the fact that none of us are doing this in a controlled environment... there are so many variables involved that strictly controlling vehicle speed won't be enough to get repeatable/reliable results; we're ballparking it to the best of our ability. But that's NOT to say it isn't an important factor... doubling your mileage isn't actually that great if it only happens at 5 mph. =)
__________________
|