Although I suppose it is conceivable that there may be some rare exceptions for which doing that would provide results that are somewhat reasonable, perhaps such as certain older Apogee speakers, and the Infinity Kappa 9, that have extremely low impedances.I'd differ with you on this point and say that connecting a loudspeaker between two transformer taps that weren't designed to have a loudspeaker between is a categorically Bad Move. For most iron-coupled amps this means that you should ALWAYS use the ground terminal for the negative speaker lead, and in the case of the Mac amps that are balanced-bridged (i.e. the MC402) to use a PAIR of speaker leads to a given PAIR of output terminals, and not to mix 'n' match a given pair of cables between taps.
The reason for this that if you don't use the assigned ground terminal, then part of the transformer secondary becomes unloaded, thus undampened . . . but is still in the feedback loop. The resulting change in the transformer's response (usually a HF peak and resulting phase lag) is then presented to the feedback loop at the opposite polarity of what was anticipated by the designer.
In the case of the more recent autoformer-coupled Mac amps, there is no conventional series output inductor, and several "feedback" connections from the output autoformer. IIRC there are usually one or two Zobel networks across combinations of the windings to compensate for inductive loads (HF damping), and additional feedback taken from 2- or 4-ohm taps that forms and additional pole/zero pair above the audioband. So to connect a speaker to disparate terminals means that the open-loop amplifier response will definitely be quite different from what the designer were thinking.
And while the amplifier probably won't break out in sustained oscillation, it will at least have quite an unwelcome effect on the amplifier's transient response, and easily wreck much of the hard work they did when they designed the output autoformer and its associated compensation network.
