Craig Brougher's interesting note has a couple of points that made
me smile, and provokes a few thoughts that may be of use in this
discussion.

First, Craig states that it's necessary to have a significant suction
difference between Theme and Accompaniment levels of the same coding
number to make the theme stand out in rolls.  That's obviously not
true: the Theme uses higher dynamic coding to make it stand out!

What's important is the base from which this higher coding was intended
to build.  My British Weber 6' grand has equal-strength springs, and
was first set up so that both regulators were identical in every
respect, moving the regulators precisely the same distance for the same
coding level.  The springs were then set for zero level, and finally
the Theme regulator was made to rise a fraction of an inch further at
zero than the Accompaniment.  The curves therefore remain a small
distance apart the whole way through, and the theme stands out very
clearly indeed when rolls are played!

Thinking about this some more: Having the levels close like this gives
the possibility of subtle interleafing of theme and accompaniment
lines.  Conversely, since rolls don't code Theme with lower levels than
Accompaniment, widening the gap between Theme and Accompaniment curves
reduces the potential of subtle shadings.  Maybe having the curves half
an inch water gauge apart the whole way gives inaudible degrees of
subtlety, but having Theme (say) 50% higher than Accompaniment for the
same level is clearly going to be unsubtle at best.

If Aeolian had intended the Theme regulator to be vastly higher than
the Accompaniment one, they would have coded rolls where Theme codes
were sometimes numerically lower than Accompaniment ones.  The fact
that they didn't (excepting Theme accents at zero Theme level) shows
that the curves were intended to be pretty close.  As, of course, do
the instructions in the regulation guide with their unfortunately
imprecise 'one degree louder' - which I take to mean sounding a little
louder.

Second, there is possibility for a slight non-linearity in British
Duo-Art boxes (and presumably American ones as well).  If the arms
connected to the rotating shaft that transmit the accordion movement to
the knife valve are adjusted to different angular positions, they will
introduce a non-linearity (the difference between two sine curves).  If
the two arms are at the same angle, the movement inside and out is the
same and the sine waves cancel to give linear response.  I think from
the markings on the shaft of my piano that having the arms equal is the
correct adjustment, although any non-linearity is in practice
inaudible.  I used a ruler to set all four arms on the two shafts to
the same position to ensure that the two sides of the box were
identical.

There is another source of non linearity in the changing shape of
the regulator as it closes affecting the relationship between spring
tension and suction level.  Of course, non-linearity in a Duo-Art box
should not come from the spring, as Craig so clearly describes.

Third: 'missing' crash valves explained away by inferior housing
stock -- this is delightfully inventive but quite implausible!  In
1924 my Weber would have cost some two or three times a new suburban
semi-detached house (UKP850 for the piano, perhaps 200 to 400 for a
suburban house), and about the same as a whole row of terraced houses.

No, these pianos were originally for the very rich who had large houses
with big rooms, the sort of houses we dream of now but can't afford.
The crash valve was not added because those who determined these things
in the UK simply didn't want them.  Likewise, according to the records,
Creary Woods.

Julian Dyer