Thanks for your comments; please forgive the chaos of what follows as I have to rush!
Agreed on all counts - the dc bias point would need adjustment and monitoring; I supposse it depends on how stable the electricity is from the wall; another story elsewhere on the site...
Regarding the CCS, the best engineered I've seen is Gary Pimm's stuff - he's made many careful measurements of various configurations and ensured that the CCS maintain their impedance at HF, and are stable. There are some details here:
(thankfully, as his orioginal pages are no longer active)
Simpler alternatives include a simple bipolar - basically a voltage reference is applied to the base, giving Vref-0.6V at the emitter; this maintains a constant current through the emitter. Performance is improved by cascading.
That said, I do like your idea of using a pentode. The idea I had was to use a pentode or something like a triode connected D3a, put a solid-state CCS in the cathode, and bias up the grid with a battery to give the CCS in the cathode enought headroom.
This would multiplies up the impedance of the CCS and also shields it from a large voltage swing (down by the mu of the triode in the latter case). Ie we use the tube as the upper element of a cascode, to fix the voltage across the CCS.
Why do I feel this is important?
Well, say we decide to go down the route of buffering the level shifter with a cathode follower, and so use a large 100K resistor in the upper leg of our level shifter in order not to draw too much current through the 6e5P anode load.
If we have a poor CCS with impedance varying from 10M to 20M with signal, then the "gain" of the shifter will vary from 0.99 to 0.995. This is about 0.5% of nasty "non-linear" solid state distortion.
To fix this, we need either a very high impedance CCS, or one where the impedance is constant with respect to level. Given sufficiently high CCS impedance, we could shunt it with a 1M resistor. This would give an attenuation of roughly 0.9 but take out a lot of the variability (if present) in the CCS.
It may be that a well designed pentode CCS is good enough - especially if you bias up the grid so you can use a higher cathode resistor (the dynamic impedance is then the gain x cathode resistance + plate impedance). My conjecture is that we could improve the performance by replacing the cathode resistor with a solid state CCS, which would be shielded from large voltage swings by the pentode; less nasty non-linear solid state artifacts. The remaining impedance would be the anode-grid capacitance (via the bias battery to ground)
Of course, how this all relates to sound is another matter! I do feel that solid state is best used when the intrinsic device non-linearities are minimised by fixing current or voltage etc.
Again sorry this is rushed, let me know if anything is unclear.