Hmm, I wouldn't say >500V, more like 3kV+ for tubes?
They can still fail shorted, particularly if you melt some electrodes together. Upside is, it takes a LOT of heat to do that. You're physically heating up bulk metal, it can take seconds before grids visibly glow (at which point you can still turn it off, if you're quick), and tens of seconds before the plate glows too many shades of red/yellow.
It's harder to work with tubes, you might not be putting in as many protective features perhaps (I mean, there's no excuse with a solid state control, but this was very much the case in the olden days), but it's unlikely to explode the first time you short-circuit it.
Transistors take perhaps 100s of microseconds to fail under fault conditions, so protective features must be considered in the primary design.
As for Rds(on), SuperJunction MOSFETs scale proportionally -- no need to prefer lower voltage types, the only reason you need stacks here is for voltage rating. (Earlier generations had a penalty going as ~Vds^2, so high voltage types truly were much worse, and there might be reason to stack them even for fairly modest voltages.)
And needless to say, tubes don't offer nearly as low Rds(on) -- it was the case as recently as a ~decade ago that high perveance e.g. sweep tubes still slightly outdid HV MOSFETs, if you ignore the heater power and screen grid requirements, the higher gm of the FET, and whether or not the body diode is a bug or feature (which basically means for switching applications, you can use the MOSFET for voltage-fed inverters, or the tube for current-fed inverters, without adding any fiddly diodes to either*).
*But you need extra circuitry to prevent grid and screen conduction while the plate is reverse biased.
But even with such exceptions, MOSFETs are well and truly the exceptional device now.
Tim