I would like first to add that the two mentions of variable lever
linkages, such as used in steam locos (John Farmer) and the Colorado
Springs central vacuum (Art Reblitz -- wow -- I had no idea those went
back to the bellows-pump days), have the desirable feature of providing
a smooth variation in the required cranking force, and not a series of
jerks as would my suggestion of defeating (propping open) the intake
valves for part of the stroke.  However, there is still the problem of
slogging down a human cranker when a blast of full chords turns up.

I suspect the Philipps scheme (Art Reblitz) is a variant of
valve-propping and would be jerky feeling on the crank.  However, yet
another, truly simple system, with no feedback required at all, is the
Welte Orchestrion pumping system.  Here, the crank is replaced by a
cam, which opens each feeder to take in air, but does not force the
feeder closed to pump it into the windchest.  Instead, a spring or
weight gently closes the feeder depending on the wind consumption.  If
no wind is being used, the feeders stay fully open, barely moving, and
the camshaft gets a "free ride."

Again, this would feel jerky to a person cranking, but with enough
feeders it might be acceptable.  At least the jerks would be evenly
spaced around the crank rotation, as with a valve-propping scheme.
Sadly, the Welte cam system works only with single-acting, not French,
feeders.

Now, if a crank organ were designed such that roll/book speed is
independent of cranking speed, any of these schemes would work fine.
I've never seen a piano style pneumatic roll motor in an organ, but
there's a first time for everything. 

Mike Knudsen