Sometimes you know you're not going to give up on a problem, because it won't give up on you.
This one had to be amenable to logic, so as things were still iffy (even with filament bulbs) and I have 2 units I decided to strip down the one with the most "repeatable" symptoms yet again.
All the busbars and contacts had been cleaned thoroughly so I only really had soldered contacts to check - these are all on the switch unit. So I dismantled the light unit and took the switch out - with the bulbs and busbars out it just has little spring plastic catches at the sides which can be pressed very easily with a small flat screwdriver and then pops out.
Taking the switch apart, including the rocker mechanism, it's actually very simple, with a flexible metal spring bar that has an openable common connector (C) at one end and the other flips between 2 contacts (say A and B).
With the switch set to off (0) the common connector C is open, when central it's A that is connected to C and on (1) it's B. All checked and tested OK with my multimeter.
I then started looking more closely at the PCB and tracing the tracks. Once you work out that some of it is really just for the little LED's that illuminate the switch internally the rest is pretty simple.
I identified which little metal stubs came from each of A, B and C and checked continuity and switch operation (reassembled) - again all fine.
So then I started following the tracks, which generally lead to metal spring-loaded slots that the busbars plug into. And here I found a problem with broken circuits!

Scraping off tiny patches of the sealing lacquer on the tracks and testing there and at the busbar slots with a multimeter I soon decided it was the soldering of the pins (from the switch, just 3 of them) that were the problem. One in particular, which had the pretty standard layout of a pin through a hole in the circuit board soldered to a track, - wasn't actually connected at all reliably.
So I removed the solder from the dodgy pin, scraped some lacquer off and resoldered and it tested good!

I presume excessive lacquer was obstructing the area where the solder was intended to bond to the track and over time what minimal connection there was had parted.
So, if anyone else is having strange issues with the internal rear light unit that's what I'd be looking at - the soldering on the little PCB that is part of the on/off switch. I had problems with both the pins of the wiring socket and those for the on/off switch.
__________________
2003 D2 FL S8. Irish Green Pearl/Beige. Solar sunroof, auto-dim mirrors, electric rear seat functions and extended leather. Ski hatch retrofit. Aftermarket reversing camera. DVB-T and XCarlink now removed as redundant.
2016 Volvo V40 T5 Cross Country (4WD) with ALL the toys including adaptive cruise etc. etc. Osmium Grey with Blonde/Charcoal leather interior. Polestar performance "optimisation".

(Even rarer than a D2 S8!) Oh, and a brand new engine at just under 30,000 miles on the factory one!

Finally: gone, but not forgotten.....
1998 D2 PF S8. Agate Grey/Platinum. Every option (I think) except electric rear seats, Tiptronic steering wheel, ski hatch, towbar & dimming door mirrors.
e.g. Cruise control, NavPlus/TV, Bose, GSM, Xenons, Solar roof, Parking sensors, Alcantara/leather everywhere of course.

(internal dimming mirror added later)
1998 (very early) Ford Focus 1.8 Zetec; ABS/TCS, Heated screen/mirrors, Aircon. Added Auto-dim mirror, Leather seats, Trip computer, Cruise control, OEM Ford SatNav with CD changer.
And before that a lot of Rover 800s, a few oddities, a lovely Triumph Dolomite 1850HL with Overdrive and way back in my schooldays an Austin Seven aka Mini 850!