My previous posts were based on the fact that I had 'done the experiment' and 'knew the answer', but having contributed them, I started thinking a bit more about what others had posted here, and about correspondence on chain rotation in West Surrey CTC's magazine. There was obviously more to it than I had previously thought. Then came Brucey's erudite reply confirming just that. Thanks, Brucey.
I have now found my maintenance notebook that goes back to 1979 and found that I did the experiment twice, during 1980-82. I was running a 5-speed bike with 14-17-20-24-28t sprockets, the 17 giving my 'normal' gear. I made two Suntour 'Perfect 888' screw-on freewheel blocks (freehubs and their cassettes had yet to appear on the market) each last two chains, more or less the way I described previously. Firstly, with block #1, two Renolds chains lasted 2100 and 2200 miles. Secondly, with block #2, two Shimano Uniglide chains lasted 2100 and 3600 miles.
So, looking at Brucey's list of things that make skipping more likely:
"- a full tooth form on the sprockets" - certainly true of those Suntour sprockets;
"- much heavier wear on a few sprockets than others" - certainly true, and
"- use of small sprockets rather than large ones" - also true, the 14 and 17 sprockets giving most skipping, and the 24 and 28 not at all;
"- chain rollers without much slack in them" - quite possibly the case for those bushed chains, quite different from modern bushing-less ones.
My rear mechanism at the time was a Shimano 'Crane', but I have no idea, at this distance in time, whether that produced "short chain wrap" or "low chain tension from the rear mech". Maybe you know, Brucey.
We live and learn. Maybe I will give chain rotation another try using the modern kit.