I've found an answer that satisfies me, thanks to those who responded.
FLIGHT, 23 February 1985, B. R. A. Burns, technical manager of BAe's Experimental Aircraft Programme
Tailplane download
In a stable tailed aeroplane the forward c.g. limit is determined by the ability of the tailplane to produce a download to balance wing lift, which acts behind the c.g. The tail must also counteract the "no lift" pitching moment generated by wing camber, which is greatest with flaps down. (A cambered wing produces lift at zero angle of attack (AoA). At zero lift, negative AoA, the wing carries a download at the front and an upload at the rear, producing a nose-down moment.)
The c.g. limit moves forward with increasing tail size at a rate proportional to tailplane maximum download capability. For this reason some aeroplanes feature devices to augment tailplane negative lift—leading-edge blowing on the Buccaneer and slats on the Phantom are examples.
The aft c.g. limit is determined by stability, and in a stable aeroplane is located a few per cent of wing chord ahead of the neutral stability point, where c.g. and a.c. coincide. The aft c.g. limit moves rearwards with increasing tailplane size, but only at a comparatively shallow rate because its effectiveness as a stabiliser is diminished by wing downwash, typically by 50 per cent.
It is a misconception that tailed aero-planes always carry tailplane downloads. They usually do, with flaps down and at forward c.g. positions, but with flaps up at the c.g. aft, tail loads at high lift are frequently positive (up), although the tail's maximum lifting capability is rarely approached.
As our tail sizing diagram shows (Fig A), there is a steep increase in c.g. range with increasing tail size as the limits move apart. In the example illustrated, a 40 per cent increase in tail size doubles the c.g. range from 10 per cent to 20 per cent of wing chord. Also the c.g. range is located well within the wing chord, so that the trimming load on the tail (and trim drag) is small.