The LM-80 unfortunately doesn't say anything beyond the testing time. It is an accepted methodology for maintaining a standard of comparison built out of the LM-79 framework.
The TM-21 "analysis" is what provides the "lifetime" guesstimates via a least-squares approximation that spits out two constants (which are supposed to be reported in the LM-80). If one knows those constants, a projection can be plotted more easily. However, they can be roughly derived from the L90(t)-L70(t) values, I believe.
But is it reliable? That depends on how well the LED follows Boltzmann statistics. It also depends on what the standard deviation of those averages are, too. There is no guideline regarding that in the TM-21 (in fact, it is not considered) from what I can see. And then there are the surrounding hardware issues...
Ultimately, the only reliable test is the one you perform. The potential may be out there, but how well can one create the optimal, external conditions so that potential can be reached? I think it is more accurate and safer to say "the LEDs will last at least 6-10khrs". Until one achieves the 50khrs mark, it should not be used as part of marketing or at least, caveat emptor knowing those values are just 5.5-6x multipliers--sample size dependent--of the actual testing values.
Hell, if someone makes it to 20khrs on 1 set of lights, that will be worth celebrating on its own, from a technological standpoint.
