This is a question about semiconductor safe operating area, so here's a quick primer on this.
Consider this amp.
Instantaneous power dissipated by the output transistors TR7, TR9 is Voltage (Vce for bipolars, or Vds for FETs) multiplied by Current (Ic for bipolars, or Id for FETs).
Voltage is the difference between supply voltage (here, 37V) and the voltage on the load (a loudspeaker). Consider TR7, at all times the supply voltage is equal to sum of TR7 Vce plus output voltage.
If you want the same power in a 4 ohm or a 8 ohm speaker, the 4 ohm speaker will take more current, and have less voltage across it.
For example at 1W instantaneous power into the load, a 8 ohms resistor as a load will have 2.83V across it and it will draw 0.35A. For the same 1W, a 4 ohms resistor will have 2V across it, and it will draw 0.5A.
So, for a lower impedance speaker, the current is higher, and the voltage across the output transistors is also higher (because the voltage across the load is lower). This means power dissipated in the output transistors is higher.
If the load is a loudspeaker and not a resistor, it gets worse, as loudspeaker have reactive impedance, so current and voltage are not in phase. With a resistive load, maximum current corresponds to maximum output voltage, which is minimum voltage across the transistors. With a reactive load this is not the case, and the max current point can occur earlier or later in the cycle, with a higher voltage across the transistors, which increases dissipation in the output transistors.
This is a problem for SOA (safe operating area). Transistors are tiny things, even a rugged power transistor will be a very small mass of silicon, like a fraction of a gram, and dissipating a lot of power into it will raise its temperature very quickly. So the manufacturer specifies a Safe Operating Area (SOA), something like that:
SOA is a rule that says how much power the transistor can dissipate for how long, before it risks blowing. You can break the rule of course (at your own risk).
Of course, you can add more transistors in parallel so they share the work... but this costs money... or you can add protection circuits to make sure the SOA is not exceeded, but these cost money too... and money is expensive!
In fact, quite often in the audio business, a nice brochure with bullshit specifications is a much more profitable option.
This means, most likely a Class-AB discrete amp that is not meant for 24/7 PA use will not have a proper SOA protection (to save costs and complexity). If it is an audiophile amp, then it will definitely not have any protection because "it would sound bad" (translation: protection that doesn't sound bad would cost 50c extra and that's too expensive).
Well that was a long post, but my point is that if the amp is class-AB (not class D) and has been designed for 8 ohms only, then using it on 4 ohms risks exceeding the SOA limits on the outputs devices, and then an output device blows, then it shorts and feeds DC to the loudspeaker, and then because a fuse in the output would have cost too much expensive money (or it would not be audiophile enough) then the loudspeaker voice coil also burns, and then you are inconvenienced.
This will not happen during normal listening because normal listening levels are very low (unless you are deaf, expect about 1W peak). However if you throw a party and let some drunk individuals near the volume button then the chance it will occur is 100%.