I agree its a lot better then the 25ºC that they achieved by only lighting the chip for a small fraction of a second and measuring its output, but why 85º and not 60º or 100º?
Good question.
I wish they had created a second binning chart at 25C to compare with the orginal XM-L’s but another guy did convert for us. 
That’s close to the temp that they are run at in real life. An LED in use stays at 25C for less than a second.
I wonder how this is determined (theoretically or by calculation maybe?), a while back i had posted a thread asking how to measure junction temperature and nobody came up with a realistic idea
Is there some sort of steady state temperature that the LED gravitates towards, because as i see it the junction temperature is a function of how much wattage is going through the led minus the heat removed, so better heat dissipation would reduce junction temperature in a steady state. So this again begs the question why 85ºC
For a car engine for the cooling system maintains a set temperature (in my car 83ºC) because the engine runs at its most efficient at this temperature, and if i read the cree datasheets properly the lower the temp the more efficient the operation.
At least for the ‘normal’ range of temperatures, an LED will be more efficient at lower temperatures. 85C is considered a ‘realistic’ temperature by CREE, and for good reason. Without active cooling, the coolest your heatsink can possibly be is room-temperature - which is almost always agreed (across many industries) to be in the range of 20-25C. At full drive, the XM-L, for instance, creates about 9W of heat. The LED chip itself has a delta T of 2.5C per watt, so even if you could keep your heatsink at room-temperature (25C), the actual die would be running about 47C. The heat-sink temperature will be determined primarily by surface area, delta-T from ambient, and air-flow. a 40C rise is considered ‘reasonable’ because it’s a good compromise between heat-sink size and temperature rise. Getting lower temps requires increasingly large heat-sinks with quickly diminishing returns for the additional size/weight.
In short, in ‘normal’ ambient conditions and without massively complicated heat management, a 25C die temperature is impossible. It makes far more sense to rate the brightness and tint of an LED in conditions that are possible and realistic, and that is why CREE switched to rating them at the realistic value of 85C. My hope is that all LED manufacturers follow suit and stop giving us all impossible numbers. The 25C rating system would be akin to an automotive manufacturer rating 0-60 times with the car starting at the top of a hill. The 0-60 time would be ‘possible’, but far from ‘realistic’ in a normal setting.
PPtk
Probably the best way to measure die temperature is by measuring forward voltage. A very quick pulse after the LED stabilizes at room temperature would give you the room-temp (25C) forward voltage. You would then turn the LED on and let it stabilize at the higher temperature and measure Vf again. DeltaV per degree C is called out in the datasheet, and you could calculate the die temperature from the shift in forward voltage.
PPtk
Interesting approach