Roster patterns
UF
I looked in to this in some detail, but about 20 years ago! The research, as I remember it now, and memory can be wrong, was thin, and in the UK was mainly in the MoD. Sectors such as nuclear and chemicals had no "clean" research. The research focussed on speed and accuracy of response when something goes wrong in complicated circumstances. Naturally, the events were "simulated", as doing real research in to firing atomic bombs and so on might upset neighbouring states.
Their conclusion was the same as the earlier poster from the military. 4 hours on, 4 hours off, with a full 8 hour break every so often - I can't remember the frequency.
However, to put against your Japanese car firm example, the UK chemical/pharma sector allows staff much freedom to set their own rotas. The staff preference was to have relatively quickly rotating patterns, with bursts of work, in either 8 or 12 hour shifts, concentrated to-gether so that time off could be bundled together. The medic.s who looked at these patterns found no particular problems, other than the "early hours" (in the morning) slower responses - but they happen whatever rotation the shift cycle uses. There were no repeats of the problems you described in the Japanese car firm. I've lots of theories as to why that might be, but they are beyond my typing skills.