I doubt frictional losses can be controlled to be a constant since rpm changes with speed, and frictional losses may not be linearly proportional to rpm. So there is no way to really isolate internal frictional losses even if you carry out all your experiments in one gear.
However there is a way around this internal frictional thing. If you go to a dyno or jack up your car so the rear wheels do not touch ground and "run" the car at say 50mph in 3rd gear, and the horsepower needed is "A". Then "A" means horsepower needed to overcome internal frictioal losses when car is operated at given condition(50mph,3rd gear)
Now you actually run your car on the road at 50mph in 3rd gear, and the horsepower needed is "B" then "B" equals to "A" plus "horsepower loss in air drag"
horsepower loss in air drag at 50mph(gear becomes irrelevent now) = B - A,
call it C.
Then use P = FV , F = P/V = C/V
drag at 50mph = "power loss in air drag at 50mph" divided by "speed 50mph". As long as you use corresponding units, they'll cancel out and give you a number CLOSE to actual air drag.
Why is the number close? because if you do the test on the dyno, the friction includes dyno and rear wheels/tire but excludes front wheel/tire friction. If you do it by jacking up the rear wheel, then friction includes rear wheel but excludes rear tire-to-ground friction, and front wheel/tire friction.