Wow, if you're pulling the "on" fet a few volts off saturation there, you're going to have troubles later -- that's going to be a lot of watts loss into heat, with attending problems.
Hopefully I'm reading that trace wrong (is it 20v or so per div)? It could be an artifact of ground not being ground everywhere (check right across one fet, source to drain), so you may have to look at that one, in this case it would be good news actually, as it would mean something other than fets ->>>> smoke.
Good news on the floating scope test! That's unusual, it must be a good one. I know you can't pull that off with older Tektronix or Flukes. Haven't even tried it with my expensive new digitals.
Yes, you probably need to increase the snubber cap till the resistor get a little warm. I pulled those values from an offline switcher I took apart recently, and from the browning of the PCB, I'd say the R got warm to get the job done. but if you're not having other issues, it may not matter too terribly much. I've gone up to .01 in extreme cases, but that burns a lot of power at high bridge voltages and these speeds. Or burns up caps if they aren't really high quality ones.
It looks like you are hitting some issues with the transformer. You may need some series L. I get it here by just shifting some of the primary to another place on the core, it's free that way.
It looks like some kind of very high Q resonance at about 3x your drive F if it can really pull those large fets out of saturation. I'm hoping that you are seeing some other artifact, but I can't know.
To do that kind of Sherlocking, I generally have to be in the room, using all my senses to see the flaw that would be obvious in hindsight. And it really is Sherlocking -- you look at everything, not just the obvious stuff that's easy to put on paper, that hair on the back of the neck stuff, that fingerprint in an odd place, whatever (speaking metaphorically of course).
Do you know the resonance of your xfrmr? This sure looks like you have a fairly low voltage secondary and a pretty high resonance (because any high volt secondary will have a low resonance, that one's hard to overcome), you might have to go higher in drive frequency? That chip gets flakey over about 80khz, however. You might want to think about changing some details of the transformer. The fewer multiplier stages you can get away with, the better. If you need more volts/stage then your diodes and caps can easily do, you can simply series them up on a per stage basis -- same number of parts total, but fewer CW stages. The bottom diodes will thank you for doing that as current will be less on them in the first stage anyway doing that. This looks for all the world like you have a resonance about 3x your drive F, which can work, but it works better with leakage L in the primary so it doesn't all reflect back to the driver at very low effective impedance. Some of of this *should* get better with a load, by the way, as it will dampen the resonance a good bit.
The fusor environment is really hard on HV feedthroughs -- the hot H is a strong chemical reducing agent, and there's also the counterintuitive Paschen's law effects, where a discharge that cannot take place over a short path, can, over a longer one. If it's any consolation, my pal at CERN and my pal at ITER really couldn't help me much on that, I had to figure it out myself.
No one else really runs these conditions, so not much has been done here -- we are truly on our own doing new technology.
And it's still not completely figured out, but what I have is the best yet, easy to fix when it does fail, and I'm still working on improvements. It comes down to this. Alumina isn't a great insulator, but it's the only thing that can take getting hit with hot D, the chemical side product evaporates, and isn't conductive anyway. But since it's a crummy insulator, you need to make much of the feedthrough out of something else, and use the alumina to take the beating, while something else does the insulating. That wire in a hollow alumina cave design that is common just doesn't cut it here -- a couple thousand dollars later, that's pretty well established as fact. Also, even things with high dielectric constant get a lot of hits while the capacitor they create is getting charged up, and even that will make them fail after a number of runs. What I'm finding (and maybe I need to post over there some more about it) is that control over conductivity is key, having some non perfect insulator here, and some much better stuff there makes it all work as you'd like, or a lot closer. For example, in my design, the outermost tubing is pyrex, which is fairly conductive as these things go. The result is that it's grounded where it passes the tank walls, and nicely and evenly divides the voltage outside the tank (no dividers or corona rings needed, it's built in) and inside the tank largely stays near ground -- since it doesn't attract fast ions, it lives pretty well. Inside that I have quartz -- 3 orders magnitude better insulator, so that's pretty important, and it has a low dielectric constant to boot (but teflon would be better if it would take the conditions). The thin gap between the quartz and the inside and outside is important too -- prevents long-path arcs down the length of the thing, that pesky Paschen stuff, some surface effect I'm guessing. It is a tight fit on the main conductor, which in my case is a 3/8" copper or Al rod (depends on which one, I've made a few) that helps take out heat in the bargain. At the very end, I have the thick alumina over that rod instead of the quartz, which takes a beating, but it's only an inch and a half long piece (goes up inside the outer pyrex some instead of the inner quartz tube). So when it fails...no big deal, it's a few pennies to replace it -- I cut it with a diamond wheel in a toolpost grinder on my lathe, but I suppose there are other ways to get that done -- the stuff is hard to cut, and the old score and break doesn't work well with it (or quartz either).
One thing I've found (hopefully I save you some sorrow here) is that a lot of layers, other than needed because you need the different properties, isn't that great an idea. If you stand off 50kv with ten 5kv insulators, it's not going to work. One will punch through, and the rest will cascade fail immediately. Methinks that Kohl book would do you some good with this, worth it at the high price, even.
This is a real problem -- like you said, we're putting the sun (actually, something a lot nastier in many ways) in a box (hey the box for the sun is space, remarkable resilient stuff), so we have to learn to build the box. And since I'm about to go to the 100kv+ range, I suspect I'll have to do more box building myself soon. Right now, all this gets into the tank with an O ring compression tubing coupler, and it's looking like I'll need larger diameter than the 1" I'm using now, sigh -- those things, and big tubing, aren't inexpensive.