Sennheiser, but I have heard and very much liked their HD580 model, which just might be the best choice for the tube enthusiast, as it represents a much easier load impedance (300 ohms); the 32 ohms from the Grado is just brutal for tube based circuits.
Now for your circuit. When designing a power amplifier, design backwards. (And, yes, driving a 32 ohm load requires a power amplifier.) Start with load first. How much voltage? How much current?
The Grado headphones are fairly efficient: 1 volt will do for most listeners. Now 1 volt into a 32 ohm load means a peak current of 30 mA will be needed to drive these headphones. In terms of watts, 30 mA is miniscule, just 14.4 mW (W = IČRL / 2), but in terms of the current limits of most small signal tube, such as the 12AX7, it is huge, as for example 15 12AX7s would be needed to deliver that much idle current. Because the 6922 is much more robust than the 12AX7, many less will be needed, 12 less in fact. Three triode sections of a 6922 placed in parallel can continuously conduct a total of 30 mA, without excessively shorting their life. Without feedback, the Zo will be 30 ohms.
The catch-22 of triodes is that if we increase the current demands on the tube, we must also increase the cathode-to-plate voltage so the greater current flow can be realized. When both the voltage and the current go up in value, so too will the dissipation. A sane limit would be 100 volts across these output triodes.
The circuit was designed for use with only a single polarity power supply, as a negative supply would be difficult to build without solid-state rectifiers. It also uses a noise canceling technique of not shunting the second grid of the Common Cathode stage to ground through a capacitor; instead, the ratio of power supply noise that makes it to this grid, when amplified at the plate, nulls the noise at the plate. The zener diodes are at the output to protect the headphones from excessive voltage swings at turn-on and turn-off.