A few quick comments:
1) The data at the start of the thread is fine, but the comparison is not -= you end up cpmparing apples and oranges. What needs to be done is a heat generated vs power generated graph for 4 and 8 ohms, with a limit drawn in for a typical limit on maximum power dissipation and maximum current - just pick an amp for an example.
2) Unless fully regulated, power supply rails drop under load, and the voltage actually varies with a frequency which is a mix of full wave rectified mains, and half-wave rectified output (talking class AB/B amps here, which pretty much covers 99% of all of them). This makes instantaneous power dissipation quite difficult to calculate. Also, as a result, the transistor generated heat is reduced, at the expense of increased power transformer and rectifier heat.
Although this complicates matters, it is important to mention because it has relevance to real world equipment. In fact, any amp that has less than twice the 8 ohm power rating at 4 ohms, is likely to make use of this fact.
Many manufacturers deliberately make the power supply 'soft', i.e. the power rail voltages are made to 'sag' under load (by careful selection of transformer size and winding, and filter capacitors), in order to keep heat generation down for complex/low loads. Some include a 8/4 ohm switch, which selects different taps on the power transformer. Often, the difference in rail voltage between the two will be small, AT IDLE. What must be taken nto account is that the fuilter caps in the power supply stay the same, so for (theoretically) double increase in output current with a 4 ohm load WRT an 8 ohm load, the rail voltage 'sags' more. This is taken into account when selecting the alternate rail voltage for 4 ohm operation, in order to keep the output power the same.
3) Designing around a transistor SOA (=Safe Operating Area) is a very difficult endevour. In theory, were it not for a phenomenon called 'second breakdown' (and indeed MOSFETs and VFETs don't have this phenomenon), one could use a transistor as long as its power dissipation is less or equal to whatever heats the heatsink to some maximum permissible temperature. If the heatsink was 'infinite', then this power limit would be the one found in transistor data sheets. In reality, the figure is 'derated' using the heatsink characteristics. It is also worth noting that because SS amps are typically symetrical, having at least one transistor that operates during half of each period of the output signal, the actual heat generated by the amp (and therefore power dissipation) is divided amongst two transistors.
A few details regarding the SOA:
Datasheet maximum current is guiven as a DC value. Because of SOA considerations, the actual current the transistor can pass reliably, can be lower, as well as higher! In general, the shorter the time of the current 'peak', and the lower the temperature, and the voltage across the transistor, the higher the current can be. It should be obvious that all of this points to instantaneous heat generation in the transistor - as current through transistor, times voltage across transitor equals dissipated power (= what produces the heat). Secondary breakdown in BJTs additionally reduces this, depending on voltage across the transistor, temperature, and duration.
4) Not all 8 ohm or 4 ohm loads are the same! Purely resistive (non-reactive) loads are the easyest to drive. For all the rest, the relationship of current and voltage called 'ohms law' applies, but with a lag, or lead. What this means in practise, is that at 0V out, there may not be 0A current. Because of this, at least one of the output transistor(s) in the amp has more voltage over it at a given current, than would be the case for a purely resistove load, and more heat is generated. Less reactive impedances of a given magnitude, are less difficult to drive.
5) Series connected speakers: BIG NO, unless the speakers to be connected in series are EXACTLY the same. Impedance variations with frequency woudl otherwise produce uneven division of power depending on frequency, and usually completely screw up the frequency response.