Leo, don't mean to detract from your point, but I'm just curious. I've run countless reliability tests on electronic components and analyzed more results than I care to think about, but have not done so for mechanical parts. With electronic components we'd use high operating hours and large samples for statistical validity (actually used component hours), but the tests were also conducted under highly accelerated conditions, because that was the only way we could achieve the necessary equivalent working hours.
Using various models, we'd extrapolate the (accelerated) test results to reliability under normal or expected conditions of use. Obviously, this would be a much higher number. Do you know if the 2,000,000 cycles used for testing engine parts was under accelerated conditions? If not, it would not appear to be very much testing per se.
Of course, conditions under the hood of a car are somewhat extreme anyway, but those are real life, rather than accelerated.
FWIW I recall being involved with prototype components being tested in road trials. They put our components on limos and taxis to get the miles they needed. Limos can rack up 2-3,000,000 miles in a year.