Under normal circumstances the grid input impedance is high, so the circuit behaviour is determined by the output impedance of the previous stage and the grid bias resistor. When overdriving causes grid current to flow the only resistance is the previous stage o/p Z in series with the grid stopper. Unless the grid stopper is large this circuit can charge the coupling cap quite quickly but it then has to discharge slowly through the grid resistor. If you want to avoid blocking you need a small grid resistor and a large grid stopper, so the time constant hardly changes as grid current flows. This would cause the voltage attenuation you describe, so is not usually done.
If NFB is present the situation can get more complicated because the previous stage might cutoff, so its output impedance shoots up to just the anode resistor value. This might help, because it augments the grid stopper during clipping.
If you want to avoid blocking, then either avoid overdriving or use a proper AB2 driving arrangement.
But I have some issues with the analysis:
(1) The current is literally flowing out of the tube when there is grid current. It can go in TWO directions: into the input Cap (at least momentarily, or for a time), and into ground (or into the Negative BIAS supply circuit.
(2) Thus the previous stage Zout is not the 'only resistance'.
(3) Apparently the ability to charge the input Cap without the complimentary ability to discharge it is what you and others have been referring to as "blocking" or "blocking effect/distortion". Like a stuffed up nose, current is stifled, and resistance rises rapidly.
(4) This is only relevant if one has a blocking cap (normally to keep out DC from the previous stage). It may be a better argument for direct or resistive coupling than against class AB/B operation.
(5) the "solution" of increasing the grid-stopper resistance as a 'cure' does appear ridiculous, and frequency-dependent too.
(6) As grid current is diverted into ground instead of back into the previous stage (now blocked by a full cap), the stability of the time constant seems irrelevant, because current flowing through the grid-leak resistor to ground drastically alters the BIAS, causing a potential runaway condition.
(7) When such loss of control over the current flow occurs, even loss of driver voltage becomes irrelevant as well.
(8) I think its "not usually done" for more serious reasons than mere drive-signal attenuation.
(9) The advice to avoid overdriving (i.e., crossing the 0-bias line) or redesign the circuit is great advice, but the steps need explicit expression.
(10) I would guess that both control-loss, undesirable current and voltage changes, and runaway tubes would make proper driver/output stage interfaces mandatory.