Have a large number of tests, let the law of large numbers rule (eg let 1000 random people borrow each type of car). Have a think about whether there's some bias between the manufacturers built in and adjust your sampling accordingly. Make the data available to scrutiny.
I agree. Paying people to do comprehensive tests in a few locations will probably cost less than a single brand new car, of which they're making thousands to millions. Even a thousand test drives is not that much; 20 cars borrowed for a month is more than enough to do those.
