This is partially wrong as well. Non-coherent energy does not add like this, and is a perfectly natural phenomenon, not something which only happens in headphones. If you play two synchronized sinusoids some meters apart, you will get perfect energy doubling only along a single line in the far field which is equidistant to both sources. Any angular divergence from this line will reduce the coherence energy relative to the phase angle. Audible sound has a wavelength of about 15m to a few cm so any real channel with any real sound will experience significant total coherence attenuation as a function of the angular offset of each pairwise emitter.
You are correct that a yelling crowd will never be quieter than a single person, but there is a massive natural diminishing effect which prevents a pure log-rule volume increase.
Right, the fact that people are not synchronized sinusoids is why coherence energy is the way I describe it. If you randomly place emitters and receivers in real space, playing real signals with real bandwidth, the distribution of coherence factors between pairwise emitters is generally positive, but is also not 1.0. This is not destructive interference, it’s just just wave mechanics.