...any particular components that give better sound with better than average parts...
As 'better' sound IMO is a subjective area that varies from one person to the next, all I can say is that you will forget the need for 'better' parts after hearing this amplifier.
Being a CFA and using a robust split feedback technique this amp makes 3MHz with those 'old' BD pair. Of course bandwidth alone is not an indicator for low distortion.
The feedback node, having a low impedance load for the feedback signal and the input device itself as the subtraction node, has it's HF roll-off pole at least 100 times higher than our highest frequency of interest. This results into an ultrafast feedback mechanism which virtually eliminates possibilities of intermodulation distortion below 100KHz.
With shown input BJTs of hFE more than 500, the OLG is static 80dB, but extends to more than 20KHz, which ensures there is no 'run out of gain' anomaly at the far highs when running closed-loop, resulting into effortless HF reproduction and zero audible crossover distortion at HF. You can read reviews in the diyAudio build thread, all praise its magic on the highs. Loud when you want it, yet crystal clear.
The output impedance is tested to be in the micro-ohms range, again for the efficient feedback technique. Ensuring superb control for the speaker cone if subwoofer is in your mind. Output impedance is, again, static even at 20KHz.
Transient reproduction of this amp is excellent thanks to the high slew rate which is scalable with resistor variation in the feedback and input collector load nodes.
This amp is three stage, and keeps itself oscillation-free without any extra capacitive compensation in VAS or anywhere. Just limit the input lowpass to say 1MHz and all's fine. That said, Miller compensation might be necessary with lower value feedback resistor. The 2K2/100R config doesn't need this.
Finally the output devices are chosen so that the amp is immune to thermal runaway and allows simply two 1N4148 diodes to bias the FETs well near ideal Iq. The diodes will also protect the LATFETs at the moment of an over-curent condition in the VAS stage, reason to eliminate a resistor in the bias node. The FETs incorporate integrated zener diodes in the package which needs no external Vgs limiter on-board.
More can be said about this circuit that uses old transistors and claims to be hi-fi. But I'll stop here. If you build this you will know the rest.