If you people would read the measurements section of Stereophile, you would see that I’m correct.
Look at the frequency response for the Moon 861. Remember, we are talking power amps here, so no preamp functions (e.g., loudness) enter into the discussion.
And the discussion:
The output impedance in stereo mode was extremely low, at 0.008 ohms at 20Hz and 1kHz, rising slightly to 0.03 ohms at 20kHz. As the two output stages are in series in mono mode, the output impedances were twice the stereo values. In both stereo and mono modes, the variation in the frequency response with our standard simulated loudspeaker (fig.1, gray trace) was negligible.
Now look at the BAT REX 500 frequency response:
And the discussion:
The REX 500 amplifier’s output impedance, including the series impedance of 6’ of spaced-pair cable, was relatively high, at 0.45 ohms at low and middle frequencies and 0.6 ohms at the top of the audioband. As a result, the variation in the frequency response with our standard simulated loudspeaker (fig.1, gray trace) was ±0.3dB.
So while the response is flat into a load resistor, it varies by ~+-0.3 dB into a real life speaker load. That’s enough of a difference to hear as a difference. And of course every speaker is going to have a different load, so the amp is going to have a different response for each speaker dependent on that load.
Know that a lot of amps show considerably more variation than this.