I've built a 2-stage Cockroft-Walton voltage multiplierwith 10nF ceramic capacitors rated at 20kV each and pairs of 12kV diodes attached in series and hooked it up to an NST with a peak output of 10kV. It worked just fine with one small problem-the capacitors began flashing when the input was around 3kV (peak).

So I immersed the multiplier circuit in transformer oil and tested it again. There was a steady arc starting from an input of about 600V all the way till I increased the input to around 5kV peak. At this point, the arc suddenly vanished and I noticed bubbles coming from the ceramic part of my first capacitor (the one connected to the supply). So I brought the input down to zero, waited for a while and gradually increased it again.

This time,however, the arc wasn't steady and increasing the input only increased the frequency with which it struck. I've since replaced the capacitor, but with the same result as before (bubbles coming off the capacitor underside, followed by a non-steady, Marx-generator style arc).

I noticed there wasn't any physical damage on the capacitors once I took them out of the oil, so I thought the capacitor leads were causing Corona discharge and decomposing the oil. I wrapped the leads of the first capacitor in insulating tape and the bubbles began popping up at the the ground capacitor. I have one last spare capacitor left and I don't want to ruin that one as well. I really want to know what's causing these bubbles and sabotaging my output. Any help would be greatly appreciated. Thanks

Corona/arcing is either boiling or outright breaking down the oil. It’s a common occurrence in oil filled capacitors and especially switches and breakers. If you get full dissociation you get hydrogen which is a much better insulating gas and responsible for how breakers and disconnects in oil work. Totally safe while submerged but if the oil level gets below the arcing it explodes. This is one way that transformer service shops spot and find low level corona, heating, and burning issues. Sometimes you have to just live with it. I can’t tell you how much is excessive and/or acceptable because I don’t work much around oil filled equipment. Most of the time adding glastic or glass or similar oil resistant materials that don’t swell can cure it. You are obviously below a Power arc voltage since your equipment survived.

Not arc flash safety or event as we associate it but interesting to hear.

I learned approximately 40 years ago in a college lab class, that some capacitors are polarized and when hooked up backwards will start to boil liquid insulator out of the attaching rivet holes. I never did hear a good explanation of why a liquid insulator might act like a diode between foil layers.

(Comical to consider my assigned lab partner was the perfect student with perfect grades. I believe he was a bit mortified by my approach, of reading the lab instructions from the back first, to hopefully discover the results we were looking for as quick as possible).

That's different. Those are electrolytic capacitors. The insulator is actually aluminum oxide. The conductor is aluminum. The electrolyte is basically glycol. So reverse the voltage and the redox reaction reverses and...boom! Also happens after you let them sit for a couple years if you don't slowly bring the voltage up in a process called reforming. Adding tantalum or using true oil filled capacitors avoids this issue entirely. And I think somehow it has gotten a lot better because I haven't had any recent catastrophic failures of capacitors that I've had to investigate unlike back in the 80's and 90's when it was a routine thing.
https://www.nmr.mgh.harvard.edu/~reese/electrolytics/
http://www.mouser.com/pdfdocs/ELNARelia ... ecCaps.pdf

There's a lot more where that came from in IEEE peer reviewed journals (subscriptions required) as well as other sources. It's a common problem that has been greatly improved over time.

Even without electrolytics, you still get failures from time to time. We had a 12.5 kV snubber made with oil filled capacitors (Toshiba brand) roughly 12"x12"x36" burst about 6 months ago in our shop on a motor test stand. It blew out all the panels, hit the shop manager in the head with a flung rivet, burned out a large tap changer, and lots of other smoke damage. it's very, very ugly. This is more "arc flash" although technically I believe it was probably an oil mist that turned into sort of a fuel air bomb when it ignited off the 12.5 kV or off just outright heating in the arcing in the capacitor. The obvious conclusion here is overvoltage took it out but we can't seem to locate a source.