I'm going to call this little beginner project a success. At least by my low standards.
If I understand correctly (I'm explaining this to myself more than anyone else
)...
When the load on each rail is equal, the circuit basically just keeps the virtual GND point steady, neither sourcing nor sinking current.
In that case, the upstream PSU (e.g. my bench power supply) will provide all the power to both loads, and very little power will be dissipated in my Darlington circuit.
But when one rail is subject to a greater load than the other, the op-amp compensates for this imbalance and begins to source or sink current accordingly and keeps VGND as well as it can.
So I think the main limiting factor to how hard I can push this design is down to what load imbalances are likely to occur in use.
Studying it with a DC load on one rail or the other and watching the Darlingtons' temperature carefully with a thermal camera, this design can sustain a load imbalance of about 2 Watts continuously. This looks quite symmetrical (+/-) and the small heat sink tops out at about 70°C at room temp after an hour or so.
The VGND holds it's set point better at higher input voltages. With an input of 30V (the maximum), and a load of 130mA on one of the output rails, that rail only sags by about 40mV, which is awesome as far as I'm concerned!
I put this together as a simple PSU project to power my future learning of op-amps and analog electronics in general, so I think it's unlikely that the rail load imbalance will get to anything like that kind of extreme (unless I connect something wrong - and that never happens obviously).
Seriously though, sincere thanks to all of you for your kind advice. After being scared of analog electronics for so long, I'm finally loving it.
Cheers!