One function of a shunt regulator is to replace the final capacitor, thus removing that capacitor's sonics from the output. It's the same function as direct coupling between stages - on the theory that no capacitor is better than even the best capacitor. The RC network maintains the low resistance output impedance, and also adjusts the response to maintain stability of the chip regulator.
The shunt regulator current is chosen to allow the full range of operation. This is an interesting design problem. For series feed circuits, the regulator current must equal or exceed the quiescent current of the audio circuit it powers. This is quite inefficient. For parallel feed, the plate choke limits the power supply current in the audio range, and we have had success using a regulator current in the range of 1/3 to 1/2 that of the audio circuit - as long as the plate choke is large enough, or the bass content is limited by an active crossover. For drivers and preamps where a current source plate load is used, the regulator current is substantially constant and a lower standing current should work - but I have not experimented to see how low we can go and maintain subjective sonics.
Presumably the shunt regulator, like any regulator, also provides isolation from line voltage fluctuations and transient current demands. You should get the same effect from a very large capacitance in the power supply, but I have not yet seen any data on what the power supply time constant must be. There is considerable evidence that it must be very long indeed; amps with absolutely huge capacitor banks are reported to sound even better when a regulator is used. Again, this is what I hear; I've not done extensive experimenting to validate it. It may be, for instance, that the capacitor sound rather than the time constant causes this observation.
