Obviously there's a range of operation from class A to class B and a circuit can fall anywhere on that range. It's interesting that with perfectly linear devices, the only truly linear modes would be class A and class B - because AB would have a step change in the waveform slope at the transition to cutoff. (one tube stops contributing). As described earlier, the increase of transconductance with current (and the cathode's ability to conduct more than twice the idle current) greatly reduces this effect, typically resulting in a distortion null at a relatively high power. And of course, negative feedback reduces distortion further.
Real circuits can have other issues, with bias point shifting with a sagging power supply, and course the real-life loadline is not the one you drew on the graph, with circuit parasitics and varying speaker impedance. The problem components are the ones that aren't on the schematic - Miller capacitance, the limited inductance of the output transformer and its leakage inductance, stray capacitance and lead inductance that can give noise coupling and unwanted positive feedback. And that damn loudspeaker that refuses to be an 8 Ohm resistor!
Real circuits can have other issues, with bias point shifting with a sagging power supply, and course the real-life loadline is not the one you drew on the graph, with circuit parasitics and varying speaker impedance. The problem components are the ones that aren't on the schematic - Miller capacitance, the limited inductance of the output transformer and its leakage inductance, stray capacitance and lead inductance that can give noise coupling and unwanted positive feedback. And that damn loudspeaker that refuses to be an 8 Ohm resistor!