After more thought I see that the two examples of servo bias I cited
earlier, one from Morgan Jones and one from Norman Koren, between
them, answer my question.

Servo-controlled fixed bias may address one or both of two issues.

First, if you have fixed bias then you need to adjust it every now
and then to allow for drift as the valves age. Of course this is not
difficult to do, and obviating the necessity may seem like an idea
for the lazy or commercially-minded. OTOH, if you really want to
forget the amp and listen to the music, it may be worth purging from
your mind the nagging question of whether it needs resetting yet.

Secondly, there is the matter of equalising the current through each
valve, which means each demands individual attention. Again, not a
problem for the ardent tinkerer, but a pain if you just want to
listen. It would be nice to set one and the others to automatically
follow.

Morgan Jones' circuit is aimed at the first, Norman Koren's at the
second. You need both circuits, or their functional equivalents, for
a full multi-valve control system. Protection circuitry would be
extra.

The equalisation part is relatively simple: you just have to set the
bias voltage of the slave valves so that the mean current of each is
the same as that through the master valve. The asymmetry of AB
operation doesn't matter because it should be the same for all. BUT,
unless valves are perfectly matched at all levels, the asymmetry
will not be same for each, and the slaves will wander a little as
the their servos hunt to compensate.

The first part, automatically compensating for drift, is tricky both
in principle and practice. Is it desirable for the system to
constantly readjust itself as the valves warm up to a constant
temperature? The objectives of the system need careful thought. In
practice there are related issues of stability (because of the
feedback and the long delay) and precision (because of the problem
of compensating for AB operation so the bias doesn't "ratchet" down
in response to the increased mean current).

Altogether it all seems like a sledgehammer to crack a walnut. And
there is the lemming syndrome.

So here's Idea #3. Microprocessor Control. Seriously. Given an
intelligent bias system, what should I want it to do, and when?

Mikkel? Another job for the remote, when it's not thinking about
anything else?

cheers, Ian