While the temperature measurement method is sort of an elegant way to estimate power loss, it is, like many things, fraught with generally unmeasured variables. And recall that the example in the beginning was just based on arbitrary quantities of the weights, time, power etc.
In addition, varying convective cooling, some radiant loss, non-constant application of input power, uneven temperature distribution within the gearbox, and so on. It can be done but a lot of stuff needs to be carefully measured to generate a valid result.