As Robert Chung has said people often believe something's impossible when they, themself, don't know how to do it. You might consider this: if you're recording power/speed data every second, a single five minute run doesn't produce one data point but rather 300.

For a five minute measurement, don't there need to be a good number of decimal points available on a speed measurement?

Here is how I see it----when it is something like aero, that doesn't change with physical conditions. if it is 80 degree or 108, the aero is the same. Same as fatigue---the aero advantage is the same no matter how tired the rider might be.

As for practical gains ... you would need to do a ride every morning at the same time and about the same temps, on the same route,probably at least six days ... one day each in each setup, three times, to get valid Practical data--as in, how much time will each helmet take off my ride time over that route..

However, if all you wanted was the difference in practical drag ... as in, in my basic aero tuck position, how much would one helmet benefit over another---that could be measured almost instantly. The aero difference is never going to change. it is a function of the shape of the helmet as it sits on your body in your tuck position. Five minutes with each would be plenty ... much less might be better, because a one-minute sprint would give all the needed data for each helmet, and you could rest half an hour between efforts to be sure each was a fresh sprint.

Generally, the precision of any of the common speed sensors (and power meters) is good enough to tease out results differing by a few percent or less depending on how many runs and the skill of the rider in holding position.

Taking the time and making the effort to see if something is really saving me time is something I don't care to do. I'm just going to trust in the Giro people who have assurred me that with the Eclispse I should be able to win the Tour de France in 2025.

For 5 minutes? No. You do need speed to be accurate, but you don't need that much precision. At least, the way I do it. Same with power data.