If you take that same 50W amp and drive it into hard clipping, the flat top and sharp edges of the resulting signal drive relatively more of the power into the high frequency ranges

__________

It's only partially correct and often over-exagerated because it's frequently quoted.

First the flat top does not contain high frequencies. Flat is associated with low frequencies.

Sharp edges indeed, contain high frequencies but their contribution is very small. If you do actual Fourier transform you'd be amazed how small the power increase in the high f domain is due to distortion. It's not even that you make a pure square wave out of a sinusoidal signals. You just clip the very top of the sin.

The danger is that by driving an amp to clipping the overall power increases and exceeds the amp rms rated power. That, in turn increases the power in high f which delivered to tweeters. The wave distortion by itself has only a small contribution.