Thanks, Ulrich!
I am really happy to own ten of these cameras. And the subject is far from being closed. But I have tried to summarize a bit that topic

It seems that these odd numbers could be due to:
- a warranty exchange. The camera doesn't work and the factory prefers the exchange rather than the repair. In that case, the original serial number is kept (Leitz did the same for the Leicas). The Fed s/n 1520 can belong to that category: the original 1a did not work and was exchanged against a 1b, some months later. Not too astonishing for an early 1a...
- an exchange with improvement, from an ordinary NKVD towards an S. The original number is kept once more (idem for the Leicas). For example, the 1b s/n 25726 which became a 1d S.
- an error of engraving. For example, the s/n 1126. A missing cipher (a "2" here) and the serial number is 1126 and not 21126, as it should be, by all its features, even the most specific ones.
- another (exciting!) possibility: a small series of numbers reserved by the factory (for the police, the army...) and made and delivered later. That could be the case of the s/n 67588 and 67610: reserved in june 1938 (serial number) and made in december 1939 (scratched date inside and lenses). For the moment, I have not found any other serial numbers between these two ones... A pity!
In fact, I am struck by the homogeneity of these cameras: they were not mounted from spare parts and each of them can be dated by its features...
If you own or know such cameras, don't hesitate to post, please. And the discussion remains open, of course.
Amitiés. Jacques.