It's sort of funny (at least, I get a huge kick out of the fact) that the least important frequencies in music are the ones that make music files so stinking big. It's the highest octaves that take up all the disk space, just for the sake of satisfying Nyquist theorem for those frequencies. Every extra octave makes the file double in size.
Because this thread is lacking in sample calculations (at least to my satisfaction), let's do one:
If we assume that we can hear from 20Hz to 20kHz, then we can hear a 10 octave range. The octaves are (in Hz):
Octave: Start: End: Nyquist: 1 20 40 80 2 40 80 160 3 80 160 320 4 160 320 640 5 320 640 1280 6 640 1280 2560 7 1280 2560 5120 8 2560 5120 10240 9 5120 10240 20480 10 10240 20480 40960
Every time another octave needs to be captured in a recording, the resulting file size doubles. For the less technically inclined (do they every visit this forum?) we call a doubling in size with a unit increase "exponential growth", which is the the mathy way of saying something gets awfully expensive in a hurry. The theoretical minimum size of an uncompressed PCM file (per second) is the [Nyquist frequency] times [the bitdepth of the recording] times [the number of channels e.g., 2 for stereo].
If we ignored the top octave, and only recorded music up to 11025 Hz (i.e., sampled at 22.05 kHz), we would cut down the file size by 50%. If the music is band limited below 11kHz, then the recording with sampling rate of 22050 Hz would still have exactly the same fidelity as a recording at 44100 Hz (or 88.2kHz, or 192 kHz, or 32 million THz, or whatever 30x super DSD is, etc.). This is throwing out 1 octave out of 10, i.e., the 10% of the range that bigshot mentioned above. Adding additional octaves beyond 20kHz extends the "musical range" by only a small fraction ("musical range" is in quotes, because nobody can even hear those frequencies) at the expense of doubling the file size for each extra inaudible octave.
Interestingly, in my samples above, you can somewhat see this effect - they're all 16 bit, 44,100Hz files, but the ones lacking the >12kHz content compress into a FLAC much better because of the missing high frequency information. The low pass filtered files are around 85-90% the size of the full files, since the high frequency information is difficult to compress. They would be even smaller if I resampled them at 25kHz or so after applying the low pass filter (which would still perfectly preserve the 0-12kHz information).