The thermal shock theory of which Lukasz speaks is not a generally accepted theory, except in the case of kovar bonded transmitting tubes with large pins. Those get gassy as discussed below. I think one could argue that the filament current inrush could be much more of an issue. That was solved by the filament system having a current limiter or constant current (which lampizator uses). I don't buy the thermal shock argument at all really. Tube's don't generally die becuase the vacuum has failed. If they do, they usually have either a purple blaze of glory (gas. Once had an awesome 5z3 that would shoot sparks at 300v, but test fine in a tube tester (around 150 volts), it looked fine) or the getter turns white. Mechanical shock (dropping a tube) is what usually causes the getter to go white. Gas happens when tubes are not operated becasue either the seal on the pins does not hold up or the internal part of the tube released too much gas from degredation and the getter (which only operates when in use) cannot degas the tube. All that said, the tubes we use do not have pins or plates big enough to drive them to get gassy. Frankly, in the hundreds of tubes I have used and tested, usually, in my experience, a tube will fail when the filament breaks or won't emit anymore (I am only taliking triode's now) because there is nothing left on the cathode. If it won't emit, it is usually becuase you have used the tube's electron material and there is just nothing left to make the cloud of electrons needed for conduction. If the filament breaks, it can happy for any number of reasons, but I believe current in rush is a much greater culprit. It's like trying to pump 10 times the blood through your body for a second. It's bad. But that has been cleared up now because of current limiting/regulation. It isn't infallable, but it stops the pop most of the time. I always turn my tubes off. The usage of the electron's on the cathode is a much greater worry to me because I have a lot more 4 pin triode tubes whose cathode is failing to emit than I have gassy tubes. I am not going to raise issues of grid emission in this, but that is another potential issue. I will also say that low voltage for some filament types (i.e., tungsten) is not bad. For others (e.g., oxide filaments) you need to be in a +/- of 5%.
Shawn