This seems to be the answer. For the metal type they called ‘Bell’ Monotype were working from types that had been newly cast by Stephenson Blake from original matrices that were made from punches cut by Richard Austin for the foundry of John Bell in the 1780s. They were used by the University Press at Cambridge in 1930 to print Stanley Morison’s monograph on John Bell. Their text size seems to be based on the original ‘English’ (about 14 point) type, which they scaled down to make the smaller sizes. (For the 8 point the descenders were greatly reduced, but the design does not seem to have been radically redrawn.) This is the ‘English’:

For 18 point and above (the metal type was cut in sizes up to 36 point) Monotype’s model was a larger type, the ‘Great Primer’ cut by Austin. This has greater contrast in the capitals and a flat foot to letter a.

For making their digital types URW’s model seems to have been Monotype’s smaller sizes, whereas for their own digital ‘Bell’ Monotype appears to have used a single model, their 18-point cut for metal.

The metal type of 1931 had been excellently made, since by then Monotype were past masters in adapting historical models to the demands of machine setting. Their Caslon of 1915 was a good example of this, in which every single size was as near as possible a facsimile of the metal types (in which all the sizes were different) cast by the Caslon foundry. Their Series 146 of 1921, called ‘Old Roman’ and later known by the US name ‘Scotch Roman’, was a similar ‘near-facsimile’, size by size, of the revived early-19th-century type (possibly also the work of Richard Austin) of the Edinburgh foundry Miller & Richard. These types have to be called ‘near-facsimiles’ since some characters neded to be slightly redrawn to fit the 18-unit system on which the Monotype line justification system depended, which sometimes meant stretching or compressing them slightly – a compromise that was rarely mentioned at the time.

Another compromise on the part of Monotype in making their metal types was to tone down some of the more strongly-drawn constrasts in stroke thickness and sometimes to adjust the design too, making it ‘smooth’ (their term), as they had done in their Baskerville and Bembo types. This technique produced the excellent fitting and even overall colour for which the ‘classic’ Monotype faces were admired. Although the ‘Bell’ is more of a near-facsimile than these faces, the Great Primer may have lost a bit of its original strength of design which can be seen in the recast types for this size that I think are still kept at Cambridge.

Maybe there is now room for a ‘Bell’ that follows the variations in design and proportion of some of the different sizes of the original type.

A very informative and useful historical insight, as always - thank you James.

"Maybe there is now room for a ‘Bell’ that follows the variations in design and proportion of some of the different sizes of the original type."

And a better digital rendition for what Stanley Morison described as "the greatest development in 'finish' since Aldus" would be much appreciated. Bell's current digital version - nor Monotype's, neither Urw's - doesn't look satisfying; it's too weak on screen and paper. I believe that someone should give Bell the same re-design treatment that Matthew Carter did on "Miller" - with an eye to its rounded and delicate terminations and serifs, which, at least to me, makes it a bit different from the "heavy" shapes of the so called Modern - Scotch types; a type more in the tradition of the French Eighteen founders à la Fournier/Vafflard.

An effort based on the display size was recently made by Paul Barnes/Christian Schwartz; but it doesn't seem to be available:

http://www.christianschwartz.com/austin.shtml

(By the way, James, I'll be in Parma in October in your seminar about type).

Thanks James, that's what I suspected.
It seems to me that there are three ways in which a metal type can change from size to size.
This is an illustration of Akzidenz Grotesk, from a 1954 specimen, with 20 pt (left) and 24 pt (right) scaled to the same cap height.

1. Letter structure: the angle of the "e" terminal is markedly different.
2. Global proportion: the x-height is different.
3. Vagary (unintentional inconsistency): the 24 pt "s" is too bold (not just an odd heavy impression, this occurs throughout the specimen).

It may well be impossible to decide whether certain inconsistencies between sizes are unintentional/mistakes, or not--perhaps things just turn out differently.

(Apologies for deviating from Bell, but this is from a scan I happened to have at hand.)