This plot of CW through a voice filter proves that the mode's efficiency is for real. At least three, and maybe four, meaningful communications are taking place within about 2 kHz here. (Time and frequency labels were left off, because they covered the lower signal.) The 400-Hz CW filters in better receivers would pick any of these right out, with any more needed "processing" being done in the operators' brains. This makes room for two or three more circuits here.
Morse "dits" and "dahs," especially on the strong CQ being sent by the 900-Hz signal in the middle, show keying transients ("key clicks") at beginning and end. At least some of this is caused by the receiver's ringy filter, and some more by the FFT math used to plot the spectrum, but the strong signal also has a rather hard keying that could use a tiny bit more shaping.
This last example comes from a European web SDR. It's a waterfall display of a small part of the 160 meter band during a frantic period in the annual CQWW DX contest. It's really impossible to count how many signals are here. Especially when considering the extreme brevity of this contest's exchanges, one comes away absolutely amazed by the amount of information being moved in only 25 kHz.
Top plot made with GRAM.EXE, bottom is a screen dump.