He was referring to transistor PA amplifiers, not tube guitar amplifiers...
It really doesn't matter: it's only a decade or two between the cherished guitar amp designs and the advent of transistor amplifiers. When those old guitar amps were designed the folks who did it were just following standard amplifier practices of the time. They didn't know that the "defects" would become a valued part of the sound.
I think it does matter.
Valve power amplifiers have low open loop gain and little, if no, negative feedback. There is plenty of scope for non-linearities, and the specifications of individual components will directly affect the amp's performance. What you call the "defects" will always be present (although they will be increased at higher power levels).
Transistor power amplifiers use op-amp techniques - high open loop gain and lots of negative feedback - to linearise the transfer function. The design overcomes the shortcomings of individual components. At low power levels, a well-designed transistor amplifier will be almost perfect. It's only at higher power levels that "defects" will become evident, due to, as I said before, the necessary compromises required in real-world design.
A simple example is frequency response. Variations in the frequency response of components in the signal path of a valve amp directly affect the response of the overall amplifier. With a transistor amp, it doesn't matter. Provided that the open-loop gain is much higher than the closed loop gain throughout the bandwidth, the negative feedback will compensate for any variations and the overall response will be as the designer intended.
What I am saying is that if you took two valve amplifiers, of different topologies, but the same overall specification, and evaluated them at, say 10% of rated power, you would find measurable differences (eg frequency response and distortion) that were audible under blind testing. If you repeated the exercise with different transistor power amplifiers, the differences would be far less, maybe even unmeasurable, and blind testing would be unlikely to reveal any audible differences.