Some more measurements last night.
With a 25mV (RMS) input, the THD is around 0.6% with a SNR of about 65dB.
The THD remains almost constant to 100mV input. At 175mV the THD rises to 2.5%, 250mV it's 5.5%, and at 500mV it is 11%.
Measurements were almost constant from 200Hz to 10kHz. Interestingly, THD falls but the noise floor rises (resulting in a lower SNR) at 20kHz and beyond.
These results are obtained with the valve operating open loop.
Interestingly, the valve bias point affects THD fairly strongly. It was previously noted that it affects the gain and bandwidth, and I'm unsure if this is an additional effect or a result of the other effects.