Redbook: only 1/4 of the master tape???

So what is the ratio of loss, if that's the right word, going from say, 96kHz to 44.1K on a redbook CD?

It's not just an issue of sample rate, you also have to consider word length. Redbook CD uses a 16 bit word whereas better quality digital recording and processing is done at 24 bit or higher. The sample rate conversion process usually entails the altering of both factors. Opinions differ but it's my experience that word length reduction is more noticeable than just a reduction in sample rate. There are different algorithms for accomplishing SRC and some are clearly more transparent than others. The amount and type of dithering also has a major influence. It's misleading to ask what is the ratio of loss in that the data reduction does not linearly correspond to what the ear hears. Depending on the type of music, the skill of the engineer and the choice of SRC processing the conversion from 96/24 to 44/16 can be very noticeable to barely perceptible. Roughly speaking it's of the magnitude of going from analog tape to vinyl.

You certainly don't need an infinite sampling rate to capture live music let alone what's contained on a first generation tape. In theory a 192kHz/32 bit recording should have 192dB dynamic range and be completely linear out 48kHz.

Here's a really good papaer from Meridian
http://www.meridian-audio.com/w_paper/Coding2.PDF

The summary is that redbook CD can produce very good quality sound when done properly (with correct dithering in particular). The problem is not redbook CD, but rather, poor implementation of redbook CD. This stacks up with my personal experience that really well recorded CDs sound phenomenal ... much better than average LPs. It's the engineering that counts more than the medium.

In addition it is not necessary to have more than 20 bits per sample (120dB dynamic range) since there is presently no audio amplifier capable of greater than 120dB dynamic range ... more bits would simply encode noise.

The reason 24 bits are chosen is because most commercial DSPs (signal processors) accept a 24 bit wordlength, and if the source is limited to 20 bits the 4 extra bits make the sample manipulation very much simpler (no overflow).

As for sampling rate 44.1kHz is only problematic in as much as the anti-aliasing filter causes either phase or amplitude distortion. Notice that some DACs are filterless (Audio-Note) to try to remedy this. Sampling above 96kHz has no benefits for 20 bit PCM.

So if the absolute optimum is 20bit*96kHz you could argue that redbook only gives 36% of the total info. However, in practise it probably delivers more than 95% of the useful information.