Transformers don't have impedances, they have impedance RATIOs. This is an important distinction.
Transformers transform impedances as a pure ratio. That is, a 4400 PP to 8 ohm transformer makes any load on its secondary look like it's 550 times bigger to a tube at the primary. An 8 ohm secondary load then looks like a 4400 ohm load at the primary. It also makes a 16 ohm load look like an 8800 ohm load if you hook 16 to it, 2200 if you hook a 4 ohm load to it, and similarly for all values in between. Power tubes have a power output that depends on matching - that is, they have sweet spot load that they do best on, most power out, and other loads will get less power because the tube itself limits how much power it will transfer out. [Actually there are two sweet spots, one for highest power, one for lowest distortion; the two spots are not the same for any known tube. From zero ohms loading up to some ill-defined number of ohms higher than the optimum power load, power tubes do not destroy themselves, they merely change how much they transfer to the load. So - if you have a tube amp with a tap for 8 ohms, you will get the nominal power of the amp only with a "matched" 8 ohm load. If you hook 16 ohms there, the power tubes "see" a proportionately higher impedance on their plates, and can only put out about half the nominal power. If you hook up a 4 ohm load to the 8 ohm tap, the power tubes "see" a load about half of the matched one, and again will put out only about half of the nominal power. This "half the nominal" power is not fixed because of the 2:1 change in load, but varies from amp to amp and tube to tube, and may not be exactly 2:1. In addition, speakers are NOT single impedance loads. It is convenient to think of "8-ohm" speakers, but the plain fact is that the speaker's impedance varies with frequency and also with the acoustic loading (cabinet and other things) that the speaker sees. That impedance meter is not going to be a huge help, because you have to specify the frequency being tested as well as the impedance to have something meaningful.
The power tubes simply refuse to put out all that much more current with a lower-impedance load, so death by overheating with a too-low load is all but impossible - not totally out of the question but extremely unlikely. The power tubes simply get into a loading range where their output power goes down from the mismatched load. At 2:1 lower-than-matched load is not unreasonable at all.
If you do too high a load, the power tubes still limit what they put out, but a second order effect becomes important.
There is magnetic leakage from primary to secondary and between both half-primaries to each other. When the current in the primary is driven to be discontinuous, you get inductive kickback from the leakage inductances in the form of a voltage spike.
This voltage spike can punch through insulation or flash over sockets, and the spike is sitting on top of B+, so it's got a head start for a flashover to ground. If the punchthrough was one time, it wouldn't be a problem, but the burning residues inside the transformer make punchthrough easier at the same point on the next cycle, and eventually erode the insulation to make a conductive path between layers. The sound goes south, and with an intermittent short you can get a permanent short, or the wire can burn though to give you an open there, and now you have a dead transformer.
For a poorly designed (high leakage, poor coupling, not well insulated or potted) transformer, 2:1 may well be marginal. Worse, if you have an intermittent contact in the path to the speaker, you will introduce transients that are sharper and hence cause higher voltages. In that light, the speaker impedance selector switch could kill OT's if two ways - if it's a break befor make, the transients cause punch through; if it's a make before break, the OT is intermittently shorted and the higher currents cause burns on the switch that eventually make it into a break before make. Turning the speaker impedance selector with an amp running is something I would not chance, not once.
For why Marshalls are extra sensitive, could be the transformer design, could be that selector switch. I personally would not worry too much about a 2:1 mismatch too low, but I might not do a mismatch high on Marshalls with the observed data that they are not all that sturdy under that load. In that light, pulling two tubes and leaving the impedance switch alone might not be too bad, as the remaining tubes are running into a too-low rather than too-high load.
