I finally got access to the TekScopes group. The x-ray images are publicly accessible here:
http://pp5ms.com/tek_hv/1.jpg (it goes up to 9.jpg)
I also found a long discussion about building a replacement multiplier from discrete components. And there is like a 30 post discussion about whether it's a "tripler" or a x6 multiplier. It was really confusing. So I watched carefully EEvblog #469 "Voltage Multipliers" and I finally understood the circuit. On the x-ray photos we clearly see 6 capacitors and 6 diodes. Therefore the circuit is a
three-stage Cockroft-Walton multiplier. It multiplies the input peak voltage by 6. The input voltage is 2.45 kV peak. Hence the HV output is 14.7 kV which agrees with the service manual. Also we see in Dave's video that each stage gets the previous stage DC voltage and the input AC voltage riding on that DC. So each stage has to withstand the input peak-to-peak voltage around 4.5 kV which is 10 kV with a 2x security margin. That explains why 12 kV diodes and 10 kV caps were used.
I tried to disassemble my dead multiplier to make sure it was the same. I confirm that it's hopeless trying to remove the potting. Acetone and IPA didn't work, I don't have toluene. But it really looks like a polymerized kind of stuff that is almost inert once it settled. I tried drilling it and even at low speed it got so hot that it started burning. Desperate, I smashed the thing with a hammer and it took a lot of hammering to see something.
My HV multiplier was a little different from the one on the x-ray photos. The arrangement is the same, but the capacitors are bigger so that the circuit occupies the whole box. I managed to measure one of the capacitors, it just had a small corned chipped so I could access both plates. They are 1800 pF. Also I found a 470k carbon powder series resistor before the HV output. We can see the same resistor on the x-ray photos, it appears as a gap in the output line.
There doesn't seem to be any replacement part available right now, so I guess I will have to build my own multiplier. rqsall pointed me to the ebay store "high-voltage-hv" which has 100 nS 12 kV diodes and 10 kV caps. Do you think 100 nS recovery will be enough for this 50 kHz multiplier ?
Regarding the output resistor, why do you think they used a regular resistor instead of a HV one ? Does it have to be carbon powder or can I use a metal film as well ? I found those on Farnell :
"OHMITE SM108035003FE RESISTANCE 20KV 500K 1%" 6 euros piece
"OHMITE RESISTANCE COMP CARBON 470K 0.25W 5%" pretty cheap
Any suggestions about the potting compound ? I don't think paraffin is a good idea, the scope gets pretty hot. Maybe silicone or polyurethane ? something that can be removed later without destroying everything.