You will never see testing data done to the extent that it would need to be done in order to be meaningful. Reasonable people understand this, but then reasonable people don't worry about discrediting a list of features.
In order to test for failure rates, you would have to buy at least 1,000 of each model, bought randomly at various shops throughout the country to avoid doping. Let's say there's 10 brands, that's 10,000 guns, and $10,000+/-. Then you have to shoot them, with at least 10,000 rounds/gun to get past the threshold of failure point for almost all the parts in an M4, and you'd have to do it relatively slow on semi-auto in order to simulate real world usage. That's 10,000 guns fired 10,000 times, equaling 100,000,000 rounds. At $.30/ round that's $30,000,000. That's also nearly 56,000 hours of test firing, not accounting for loading and cooling, at 2 seconds per shot. That's 7,000 eight hour days.
Even if you reduced all of this by a factor of 10, and only bought 100 examples of each gun you would still take YEARS of firing alone just to complete the test.