perhaps i am late or do not know of such a codec, but ... how about one:

:where you decide what frequencies are to be encoded resulting in a 'semi-lossless' file using a band pass filter with (you have option) : 'brick walls' at both ends; OR a cutoff and high cutoff with specified higher order filters of bessel, butterworth, Chebychev 1, or Chebychev 2.

this of course is the main feature. these options would be ideal for someone who wants to encode a wav to a file that would have only those frequencies that they can hear. so if they can hear only 100 - 10khz frequencies, they can choose to only encode to have (99.99-99%?) the sound quality of the music in that frequency range.

:supports 8, 16, 20, 24, and 32 bit depths

:supports ANY sample rate resolution. i mean not just 44.1 and 48 and their (derivatives? or wrong word im choosing...). i mean you can enter 200khz or whatever you wish. i say 200khz cuz lynxTwo supports that. i guess a limit of 1mhz and minimum of 1khz would be ok?

:respectable compression ratio

:at least 7.5x real-time encoding and 5x real-time decoding speed, though i assume it would be better than that.

:compatible with at least m$ os (xp,longhorn)

:and of course support tagging / metadata, either proprietary or an established existing one

:an excellent supplied gui/frontend, community support, and ppl who can make popular apps utilize it (ex:foobar,eac,etc) ... so it can have staying power!

You can just apply the filters and then use a lossless codec. Of course you'll loose most advantages of using a lossless codec, but if this somehow suits you, then by all means go ahead. You might even be able to find some commandline utility to apply the filter you want.
And yes, this should usually result in some space savings, as filtering out high frequencies will make a signal easier to predict.

You could use sox as a commandline utility to apply low/highpass filters.