Kevin: The funny thing is that before I built my 6AS7G cathode follower headphone amp, I also considered internal sparks/ion discharges to be fairly rare in small glass receiving tube types. In fact I only recall seeing it once more than 15 years ago, in a ham radio transmitter, where the 6146 output tubes had been severely abused.
However one of my RCA 6AS7G tubes does arc internally, and yes, I was wearing the cans when it happened. Not fun at all. The amps is not DC coupled, so no permanent damage was done, but I'd rather not repeat the experience more than I absolutely have to.
Sidebar (probably preaching for the choir here, but alas): In large ceramic/metal disc seal tubes of the 4CX1000 variety and larger, ion discharges are considered so much a part of normal operation that the surrounding circuit is designed with this in mind. Usually this is done via a large wattage 50-100 ohm resistor in the anode lead, plus an over voltage crowbar on the screen, if present. The large resistor current starves the discharge, extinguishing it, while the screen circuit discharges the over current to ground, protecting the screen supply- and bypass capacitor. Check the Eimac literature or the service manuals of your amps. I'm willing to take large bets that each and every one of your large tube amps will have this protection built in as a matter of course. End sidebar.
Subsequently I researched the matter a bit, and it would appear that the 6AS7G/6080 family is prone to this problem as well, especially the old RCA US variety. Perhaps somewhat surprisingly Atma-Sphere et al. suggests that current production Sovtek/Svetlana tubes are the best and most stable 6AS7G type tubes available anywhere. But perfect they are not. One Atma-Sphere customer
gives non-documented figures of around 20% of new Svetlana 6AS7G suffering from problems in one form or another. He suggests pre-screening the tubes, before they go into an amp. Notice his note about sparking tubes and not having the speakers connected while testing and burning in new tubes.
My thinking is to bridge each output with a series/anti-parallel combination of high speed/high power diodes, hard limiting the output voltage to, say, +/- 1.4V. Any type of relay protection would not be fast enough by orders of magnitude to stop the discharge, so we might as well add some sacrificial lambs, who can take the kick from our supply caps. The protection circuit you describe could then shut the amp down some time later if the problem persists.
Add two similar diodes across each grid/cathode, limiting the grid voltage to max. 0.7V positive, and the driver circuit might not mind a spike too much. My hope is that the non-linear capacitance of the non-conducting diodes are so low, that it doesn't matter for the signal quality.
----
I have been speculating: What would be the trade-off if you limited the quiescent current to what a single 6AS7G in each branch could cope with, say 100mA or so, assuming the amp was only intended to be used with cans with 100 ohm or higher impedance? If you believe it would be worthwhile to scale the circuit down to 4x 6AS7G @ 2x 100V or perhaps even as low as 60V anode voltage, then I would be quite interested in building one of these monsters in the near future. Finding the semiconductors will be fun, seems my usual suppliers doesn't stock most of them...
Given a lower current a stabilized PSU could be made via a pair of suitable shunt (tube?) regulators for each channel. They wouldn't need to secure the DC OP, so the power consumption could be designed to be within 'reasonable' limits.
I would also consider using a PCB, one per channel and including safety circuits, but with the two tubes off board to give me greater freedom when it comes to the physical layout.
