There is some accurate and some pretty iffy LED info in this thread. Since I sometimes work with high powered LED technology in my industry (displays, not aquariums), I'd like to clarify a few points.
1. When the LED mfrs state a lifetime of 50,000 hours or so, that is a theoretical lifetime based on the known physics of LEDs. It is not a tested lifetime. Since it would take about 6 years to run that test, no one including the LED manufacturer has typically run one that long before initially releasing the part for sale. This estimated lifetime is generally considered to be pretty accurate under the ideal operating conditions that the lifetime is stated.
2. Achieving that rated lifetime in a light fixture requires that the lighting designers adhere to certain optimal junction temperatures of the LED die. These junction temperatures are not generally the same as the maximum junction temperature that is specified. A maximum junction temperature might be spec'd at 150C, but stated lifetime performance might be achieved at something under 100C. I saw one aquarium light mfr state that as long as the max 150C junction temperature wasn't exceeded the fixture would operate fine and that is simply not true. When you see a life rating on a light fixture, you can bet the manufacturer is just passing along the LED life rating from the LED mfr whether or not their fixture is operating those LEDs under the conditions needed to meet those lifetime specs.
3. High junction temperatures adversely affect LEDs short term for brightness and it tends to shift the color temperature of the light output. Lumen output of the LED are typically rated at some unrealistic junction temperature such as 25C, but will drop in brightness down to perhaps as much as 60% of that value as the junction temperature increases towards maximum.
4. High junction temperatures over time shorten the life of the LED, sometimes dramatically. Lowering the junction temperature, such as by running in dimming mode will lengthen the life of the LEDs - up to a point. For instance, I know one mfr that rates their super high power LEDs at 50,000 hours, but will increase their estimated life expectancy up to 80,000 hours as junction temperature is reduced but then they hit a theoretical limit as the LEDs tend to have other wear-out mechanism besides just the junction temperature. Obviously some mfrs of lighting might take the tact of hotrodding the LEDs to get the maximum brightness possible out of them as that is what the hobbyist wants to see, but that will directly lead to a shortened life.
5. Junction temperatures are a function of how much power is being put through the LED, generally in the form of current and how much heat is being extracted from the LED in the form of heatsinking. Low thermal resistance between the LED die and the heatsink is very important. Without cooling on the heatsink, you want it to get hot as a sign that it is pulling the heat away from the LED junctions, but when you apply the cooling to the heatsink, you want it to be adequate enough to significantly drop the temperature of the heatsink so that there is a good thermal gradient between the junction and the heatsink to help pull the heat away. A heat-soaked heatsink loses efficiency. If your heatsink is cooking, so are your LEDs probably.
6. I am sure that many lighting manufacturers pay little attention to the conditions that are necessary to achieve the maximum life and stability of the LEDs in their fixtures. To really characterize the junction temperature of an LED in a system is not a trivial exercise. It is very easy to build a fixture that turns a 50,000 hour LED into a 10,000 hour LED. It is also easy to design a fixture where failure of 1 LED increases the power through the other LEDs thus having an aging/failure ripple effect. You can also design the fixture so the bad (open) LED prevents the rest of the chain from operating.
7. As for LED ballasts and the like, all mfrs publish a maximum thermal spec. That thermal spec should not be construed that if it is set at 50C (for instance), that you are 100% fine running at 49C and going to instantly fail if you run at 51C. It is much more like a sliding scale with the closer you get to the upper range, the higher the failure rate will be. The upper temp spec is often gated by safety certification agencies like UL that care about burning stuff up and not about how long things last at different temperatures. Adding additional cooling to a particular unit that is running near the upper temp limit may not help, but you can pretty much be guaranteed that if a population of the devices are kept cooler rather than being run at their upper limit, the overall failure rate will be less over time. I like to think about it like the RPM redline in a car. While the car manufacturer says that you can run the motor up to 6000RPM safely, you know darn well that if you run around at 6000RPM all day, your car isn't going to last nearly as long as it would if you cruised around at 2000 RPM.
As one hard data point on LED life, we evaluated one well known high powered LED brand where we conducted measurements of stability over time measuring both brightness and color. In this case we were testing Red/Green/Blue LEDs and no white. At 10,000 hours we had measured less that a 1% drop in brightness and almost zero color shift using lab quality equipment. In fact the temperature fluctuations in the testing room had a bigger effect on brightness than time in this test. We were able to chart each time the air conditioning kicked on in the room. These devices were operating in a pulse mode of operation vs constant on, so this provides a more optimal junction temperature for best lifetime, but my point is that LEDs can be very stable over time given the right conditions.
I will also comment that white LEDs are typically phosphor converted blue LEDs. Phosphor has a similar thermally accelerated aging curve and some of the phosphors are bound with epoxy which can also yellow with age and temperature. I would expect that white LEDs would tend to be less well behaved over time in some regards for these reasons, though I don't have much long-term experience with them.
When you factor in not only all of the manufactures who in this industry vary from fairly large companies to garage shops, but also all of the DIY'ers that are randomly combining LEDs, heatsinks, fans and thermal paste into light fixtures, it is easy to image that the exact same LED might be lucky to last 5000 hours in one setup and be good for 25,000 hours in another. The desire to get the maximum brightness possible at the lowest cost possible in this hobby will probably always overrule achieving maximum lifetime because the first two affect the buying decision almost exclusively.
