Low-order passive crossovers (6 or 12 dB/octave) are less sensitive to amplifier output impedance (damping factor) than high-order crossovers (18 or 24 dB/octave). High feedback is the most direct method of reducing output impedance for both tube and transistor amplifiers.
Which in turn affects how an amplifier clips. Unless a "soft-clip" diode array is used in the feedback loop, but those have other, unwanted side effects, such as increased distortion in the top 3 to 6 dB of the amplifier’s power curve.
It’s considered good practice to let the output devices define clipping. At that point, they are giving all they have, forward gain collapses, and feedback loses effectiveness. The transition zone from linearity to clipping might involve charge storage if solid-state devices are used, or the power supply for the driver stage could sag, and let the drivers clip at the same time. This extends the recovery time from clipping, with power supply rails fluctuating up and down as the circuit recovers.
The SET amplifiers without feedback do not hard-clip, but if the driver and output section have a common B+ supply, that supply will sag when the amp clips, and if the driver and output section are RC-coupled, then the coupling cap must re-charge before normal bias appears on the grids of the power tubes.
If regulators are used, they must tolerate overload conditions and recover quickly, preferably within milliseconds. As you can see, a lot can go wrong if the speaker demands more current or voltage than the amplifier can deliver.
I should mention speakers not only consume energy, but they reflect it back to the amplifier as well. This is what a reactive load does. The most severe condition is a pure reactance with a phase angle of 90 degrees. That reflects ALL the power back to the amplifier, which can create an overcurrent condition in the power transistors. Tubes tolerate this better than transistors because it takes a gross overload lasting several minutes to overheat the plate of the tube, while a transistor can fail in milliseconds.
It’s kind of sad how low speaker efficiency really is. We think of a 92 dB/meter/watt speaker as "efficient", but it’s only 1% conversion efficiency. (A more typical 86 dB/meter/watt speaker is 0.25% efficient.) The 1% efficient speaker needs 100 electrical watts to create 1 acoustic watt (which is very loud).
What makes it worse is the other 99 watts (which are quite expensive) do nothing but heat the voice coil, which is undesirable because it is in a very small space and is not easy to cool. A more efficient speaker has the advantage of cooler voice coils for a given playback SPL, which gives it greater headroom.
