Thanks very much to everyone who replied.  I was imagining driving the
take-up spool with a strong stepper with a servo drive (prevents missed
steps) so the take-up moved in fixed increments of angle, and putting
a moderate and consistent tension or drag on the feed spool so that the
paper wrapped around the take-up spool at a constant tension.

However, feed rollers would be nearly as easy and I already have
some good ones scavenged from discarded inkjet printers (frequently
a treasure trove of optical encoders and sensors, a variety of rollers,
and a precision ground steel bar or two).  I will look at the feed
roller approach instead, but it still leaves the subject of Tempo
compensation slightly mysterious.

I have no interest in punching multiples of the same roll, so hadn't
thought about that, but I do see why relying on the diameter of the
take-up spool wouldn't work if I did.  (This doesn't prove original
machines didn't have compensation, as presumably they could have changed
the speed of the master instead of the new rolls' paper if they'd wanted
to).  It would be easily done in software instead if it turns out to be
needed, but I'm not completely convinced that it is.

Incidentally we have access to some handy tools nowadays that were not
available to older machine builders.  It turns out that paper slip is
pretty easy to detect with an optical mouse sensor -- many are a single
type that is not hard to adapt.  I think I'd want to detect any slip for
any reason, and immediately stop.

Lester Hawksby, UK