Class B power
Imagine that we have an ideal Class B amplifier, one that can have its output pin go through rail-to-rail voltage excursions with a zero headroom requirement in each polarity and let's hook up that idealized amplifier to a ground referenced load.
We examine three parameters as a function of the output voltage to the load. We examine the power delivered to the load, we examine the power lost as heat in the amplifier itself (the power dissipation) and we examine the power efficiency.
Even though this amplifier is idealized, we can learn something about real world amplifiers from this idealized exercise and yes, the screwy looking rail voltage numbers here have been chosen with malice aforethought to make the numbers easier for discussion.
Power to the load increases, of course, as the square of the delivered output voltage. Power efficiency varies monotonically versus the delivered load voltage, but I'm not quite certain of it being exactly a linear variation, even though it visually looks that way.
However, power dissipation varies non-linearly versus changes of output voltage and actually maximizes in this example at only 64% of the maximum output voltage capability.
The cautionary tale is not to just take a Class B amplifier to its maximum output when calculating worst case power losses. Take a look too at the conditions at less than full scale output as well. There could be more power dissipation happening than you might otherwise have supposed.