Having read Ralph's article "Competing Paradigms in Amplifier and Loudspeaker Design, Test and Measurement" I have some thoughts that I should share. I agree with Ralph that the simple-minded idea that the voltage-driven speaker is always best is not a good idea, and should not be the basis for judging the quality of amplifiers. Ralph is right in questioning this long-held view propagated by review magazines over the years. It makes it easy for them because measuring is generally easier than critically listening in a controlled and scientific manner.
I have built various experimental and production amps over the years with various output impedance characteristics. On many of my production amps I have provided a feedback switch that allows the user a choice in output impedance characteristics. Some of the experimental amps have included negative output impedance, as well as current (very high output impedance) drive. Various speakers react in varying ways to these amplifiers; sometimes well and sometimes poorly. Users of my productions amps that include a feedback switch are all over the map as to what feedback setting they prefer. Some of this is personal taste and some has technical merit.
Some years ago I invented a measurement device utilizing RF technology that could be applied to the cone of a woofer to measure its movement in response to a step in voltage and to display the movement on an oscilloscope. The voltage applied was delivered from an amplifier that could have a very low output impedance, but the output impedance could be increased by either reducing feedback or by adding a series resistor to obtain a known driving source impedance. I then applied a single step to see how the speaker cone responded. The ideal response is for the cone to move to its new position as quickly as possible, but not to overshoot and ring. This is called critically damping. If it rings, it is under damped. If it takes too long to reach its new position it is over damped. Too low of a drive impedance results in overdamping and too high output impedance results in underdamping.
Of course all of the drivers in a speaker have different damping requirements and simply adjusting the output impedance of the amplifier for critically damping in the woofer may not give the best overall sound. It still comes down to critical listening folks.
Unlike Ralph, though, I would not assume that a constant power drive to the speaker is best either. First, I think a bit of clarification of constant power is needed. For an amplifier to truly deliver constant power into varying load impedances it would have to have extremely complicated feedback circuits that would need to determine the real (resistive) components of current and voltage, then make a mathematical product, and then force this product to remain proportional to the input signal.
Ralph also mentions the 4, 8 and 16 Ohm taps on the output transformer in tube amps as providing this ideal constant power. In reality, these taps are provided for impedance matching for allowing the tubes to provide the maximum power output that they can under the other constraints of the amplifier such as power supply voltage. These taps are specified for recommended speaker loading and are not the same as output impedance. An 8-ohm tap is used for a nominal 8-ohm speaker, but the output impedance at that tap may be 2 ohms. Just because a certain impedance transformation ratio can result in the highest obtainable power output, it does not mean that this ratio provides the lowest distortion. In my designs, I tend to "light load" my tubes, meaning that I use unusually high transformation ratios that are possible with my ZOTL technology. On the other hand, Ralph uses heavy loading which is required for the traditional OTL type of amplifier. Any other loading would require an unreasonable number of tubes.
An impedance-matching device such as an output transformer or my ZOTL will reduce the output impedance of the tubes as presented to the speaker. Progressively lighter loading of the tubes via the impedance-matching device will also reduce the output impedance presented to the speaker before any feedback is applied.
So what really makes sense in terms of distinctions between voltage, current, and power drive to speakers? I have already implied that voltage drive is likely to over damp the speaker, while current drive is likely to under damp the speaker. Overdamping results in overly tight sound and underdamping results in loose undercontrolled bass. If we were to forget about the reactive components of the speaker for a moment and consider only the real resistive components to which we can transfer real power, the closest thing we can come to for constant power without all of the feedback goop I mentioned earlier is to provide an output impedance that is equal to the speaker impedance. We could get quite close to constant power in the following example. Consider an amplifier with a source impedance of 6 ohms and delivering 10 Volts before the load is connected. The maximum power of 4.16 watts will be delivered to a 6 Ohm speaker and the power will be 4.0 watts into a 4 Ohm speaker and 4.08 watts into an 8 Ohm speaker. This is quite good for constant power into a purely resistive load, but unfortunately a speaker is not purely resistive.
In my own listening, I find that I like having a source impedance of about 3 ohms for many of the speakers that I use. This is a bit less than half of their nominal impedance. If I take an average of what feedback switch settings others are using with my amplifiers, my tastes here seems to be in line with what others are finding. I do think that large amounts of feedback in amplifiers should be avoided, but small amounts are not harmful and can help to get the output impedance optimized.