Offhand I'd say that the amplifier oscillated when you powered it up, and the eBay transistor, being counterfeit and only good for 20 watts, failed. Put some good transistors in there. Up the gain of the amp so it's less likely to oscillate, and try again. Also, turn the bias current up. An amp is more likely to oscillate if it's got a class B hole.

hmm, but when I tested it without the power stage, there was no oscillation. I.e. the voltage amplifier stage worked smoothly, could it be that only high power BJTs caused the oscillation?
Oh and  didn't plug any speaker, so no current was drawn from the output.

Yes, the output stage can cause it to oscillate, even with no load. You might need to add a 100 or 220pF capacitor base-collector on Q15. And as I said before, if there is no bias current it can push some amps into oscillation. Have you broken open the bad transistor? I'd be willing to bet the die is only 1.5mm wide, should be about 5mm. Fake transistors are hard to avoid when shopping online.

Thanks for the answer, I'm onto new transistors now. Could be so kind and briefly explain the relation between bias current and amp stability?

The output stage has a voltage gain of 1. With the bias circuit adjusted below 2x Vbe, the output stage has a dead zone with zero gain, and the driver stage has to quickly slew when trying to change output polarity. And the compensation network charges up when the drivers are slew limited, so when the output finally does respond, the driver stage overshoots. If your amp is overcompensated or otherwise very stable, it will just be a class-B amplifier with some crossover distortion. Otherwise it will oscillate like crazy. Having some bias current at all times insures the voltage gain of the output stage is always close to 1.

As an aside, with MOSFET transistors the situation is even worse. The relationship between gate voltage and drain current is not monotonic, there is some hysteresis around 10 to 50 mA of drain current. As a consequence, bias current in MOSFET amplifiers needs to be set higher than 50mA per transistor. But thankfully you are using bipolar transistors.

All right, I see. Dead zone will cause zero output which will through the feedback make driver circuit, i.e. VA stage, drive too high and overshoot. So when output finally shows something out, it might start oscillating. Thanks a lot for your time to explain this.
For the next step, when my transistors are replaced, I'll limit the both supply voltage and current and slowly increase them until I see if the output is stable, while adjusting bias current.