We did build a test buffer, which used 12AT7s or 12AU7s as cathode followers. I chose the cathode follower because that is the circuit most audibly sensitive to power supply performance. The test vehicle had six different shunt regulation schemes - a gas regulator tube, a string of Zeners (bypassed), and four different triode-based regulators, augmented with various bits of silicon. The two best sounding were very similar hybrid regulators using the TL431 or TL1431 chip with a triode in cascode to handle the voltage (the chip is limited to 30v maximum). We chose the simpler of the two.
We did not follow up with optimizations of each candidate - for example, a high-voltage transistor or FET could be used as the cascode device, with a suitable heat sink. We had at that time been working for many years with John Tucker and the late John "Buddha" Camille, and they had done more extensive experiments - this must have been back in the late nineties. They had concluded that FETs just didn't sound as good, though they worked very well on paper and in measurements.
Here's an interesting historical note. Some 15 years ago (before I began designing for Bottlehead) Bottlehead was using paralleled triodes for drivers, 5965s and 6N1Ps mostly. Eventually Bottlehead switched to single sections, finding a bit of smearing (for want of a precise technical term!) in the sound of paralleled triodes. That left a spare triode in each monoblock. As I recall it, I made an idle suggestion to use that triode for a shunt regulator and John Tucker immediately built one. I really did not expect it to be audible, since the driver was already isolated from the power supply by a C4S plate load. I was so astounded when I heard it, I insisted on switching amps (only on channel was regulated) and then turned the amps over on the floor to trace the circuits for the next 15 minutes. I really suspected Tucker of pulling a fast one!
Anyhow, that's how we came to use the triode as the heat- and voltage-dissipating element. There has never seemed to be sufficient reason to try to develop a different approach, beyond that initial test device to confirm we weren't missing something obvious.
