Back in '77, op-amps weren't as good as they are now in terms of PSRR and output power. That push-pull output stage is quaint but given that they wrapped it in the feedback loop of the last op-amp, it may have been a necessity at the time.
Also, the discrete output section helps power dissipation in the regulator. Especially if the output gets shorted.
... And I was curious, so I just looked at the datasheet for RC4558. Maximum supply voltage is +/- 18V. So you'd probably have to bring that unregulated "20V" rail down to 12-14V actual to stay within that.
Ah, but that section here is a feedback-stabilized amplifier stage; and the power rail is dual-voltage. The ripple in the positive rail swings opposite to the negative one.
Lots of power amplifiers use no voltage regulation for the rails. The amplifier is a kind of voltage regulator already anyway; if you add a regulator, you're basically adding another amplifier to the amplifier.
(Hey, I heard you like amplifiers, so I put an amplifier in your amplifier, ...)
But most of my experience is analog conditioning for microcontrollers - more sensitive, higher frequency noise, and admittedly single rail with today's op amp technology. An extra component doesn't necessarily add complexity like in software, but can actually simplify one's mental model of the circuit - nice stable supply rails.
If you look at that original circuit, how much did the regulator cost compared to say the transformer? Eliminating the regulator would have required changing the transformer, at least.
I'm just trying to get across that it's a bit of a red herring to prioritize cost-optimizing the circuit if you're setting out to build a one-off clone. Even hobbyist debugging time is worth more than a few ten cent parts.