It's obvious you are looking for that sweet distortion at a lower
level. Whether the reason is to avoid a visit by the cops, or that the
speakers can't handle all the output doesn't matter.

One option is a power soak. These are just banks of high wattage
resistors in a box with switches to configure them for different
amounts of attenuation. This option removes the interaction between
the tubes/output transformer and speaker(s). For the purist, this
interaction is part of the sweet sound and is important.

Of course if you have to have it all that is not good enough. You have
options. Use a lower capacity speaker system to start with. The VC and
output Xfmr interaction is part of the cone excursion of the speaker.
To the amp is looks like saturation in a way, but it's not quite the
same. Much has been devoted to this and I am not prepared to go into
seventy years of history. But decoupling the VC from the amp will not
get you the same sound. Also, running the amp at a lower power will
never saturate the output transformer core, but eh, we do the best we
can.

Actually the best way one might think is to somehow reguilate the
plate voltage down. Doesn't matter voltage or current. But there is
another way. This will get you closest to what you want. I really
don't know how it will affect the output tubes, but I can guarantee
they won't overheat.

Lower the G2 voltage. Now if the amp is grid biased you will have to
adjust that, but if it is cathode biased it will adjust itself.
However, being familiar with cathode stripping in CRTs, I can't say it
won't happen to the output tubes. I don't THINK it's a big concern,
but don't take that to the bank.

The lower G2 (screen grid) voltage makes the tube behave more like a
triode. Thus it's output impedance is higher. This allows every
abberant change in current drawn by the load to interact. You could
literally have a switch on the back. If the outputs are cathode
biased, it should work just fine. Flip it one way for low power and
the other for high. If you're concerned about cathode stripping there
are two options. One is a simple megohm resistor in series with each
G1. To get fancy you could use a diode with the anode connected to the
control grid (G1). If you use the diode put about a ten meg across it,
no tube is perfect. But you do this separately for each of the six
tubes.

I think a one meg, or even a 2.2 meg would work just fine. In fact you
can go as high as you want until the interelectrode capacitance comes
into play. Always use a scope, because it could be oscillating. You
don't want that.

I don't think cathode stripping is the issue, but the operative word
here is think. Nobody wants to wreck $100+ worth of pentodes.

JURB