One thing I've noticed is that many here in the Green Room like to compare real-world CFL lifetimes, which are noticeably shorter than advertised lifetime,
CFL lifetimes can be all over the place, depending upon the build quality of the bulb, where it's installed (near doors often means vibration, small enclosed fixtures cause high operating temps, etc) and how the power quality is in the home (some homes have lots of interference on the lines that play havoc with the tiny little circuit trying to push steady current to the electrodes, or big transients from appliances starting and stopping). From talking to various people who use them or tried them and gave up, it appears to me a small, but non-trivial minority have consistent problems with them, but most people do not.
I've not had a single CFL fail in the last 5 years, and I don't recall any failing in my previous home, either. Some of those in the common areas of my home easily exceed 3 hours per day, putting them easily over the 5,000 hours normally expected for a CFL. A couple of my LED's are probably right around 5,000 hours, too, because I put them in the fixtures that are on the most.
I expect a similar trend for LED's as CFL's. Some of them will fail early due to defects. Most should get in the ballpark of the expectation. The most likely failures would be in the power supplies or the solder joints hooking everything together, which is primarily a matter of quality control by the manufacturers. The industry has done quite a bit of testing of the actual emitters, and the failure rates on those are extremely low. They usually very slowly and predictably lose brightness as the phosphor coating breaks down over time.
When Philips won the L-Prize for one of their high-performing LED's a few years ago, they first had to pass 7,000 hours of testing at 110 degrees F, and the trends for brightness, color drift, and power consumption were extrapolated to 25,000 hours, with a requirement that less than 10% fail. Since the DOE had already set up the test, they kept it running. They finally ended the test at 40,000 hours. The lights were still producing 95% of their original rating, with zero bulbs failing in any manner. Dimming failure was considered dropping below 70%, so they had a huge margin beyond expectations.
Phillips probably cherry-picked components for their test bulbs, but that kind of margin leaves a lot of room for worse performance before bulbs are actually consistently failing.