Land-line phones pass through basically everything from about 300Hz to 3300Hz, since decades of research and usage have shown such range to be adequate for people's voices to sound the way people expect them to sound on a telephone. Digital phone systems have, for the last couple decades, sampled audio at an 8KHz rate with 8-bits per sample and sent is as a 64kbps stream of bits.
Digital mobile phones need to use lower data rates, so rather than trying to send all frequency components simultaneously, they identify the most important parts of the signal and just sent the characteristics of those. All other information in the signal is abandoned. Audio which has been processed in good enough for mobile subscribers to tolerate, but it often doesn't sound a whole lot like the original, and is basically useless for forensic analysis.
To add to what you said:
http://en.wikipedia.org/wiki/Mean_opinion_score - audible deficiencies in codec quality.
The eight bit sampling at 8 kHz with no compression example, is G.711, which already has artifacts showing up.
These are just for the basic encoding. Microphone quality, any digital to analog coversions, and recording medium all worsen this - without getting to microphone placement and background events; or decoding reconstructions due to bit loss, comfort noise,and echo cancellation.