(I know just enough about this topic to accomplish what I need to, so please forgive layman’s terms.)
I use a PC to time-shift some radio shows I like. Most of the shows are very talk-heavy; I record those at 96 Kbps because trial and error showed me that was good enough. But one show is very music-heavy, so I record it at 128 Kbps, because that’s sufficient for my needs but gives me better sound for the music.
Here’s the thing: on that music-heavy show, there is noticeably more high-end noise (you can actually hear the hiss kick in a second or two after the recording starts as the recording software’s volume-leveling setting finds its happy spot).
The way you get more compression is to get rid of frequencies that are less likely to be important. Mp3 is a ‘lossy’ format. High end frequencies are the first to get booted.
For comparison, MP3 encoding at 128 kb/s using LAME (-V 6) employs a low-pass filter at 16538 Hz – 17071 Hz, and furthermore encodes frequencies above 16 kHz less accurately in case a more accurate encoding would cause an increase in bitrate.