I believe the M-Audio problem is because they won't release specs for their gear. The UA-5 seems to work fine.
One big problem I have with linux audio is the lack of good dither. It would save me a lot of hassle if I could just 24/96>16/44 from a command line without ever visiting windows. Audio and Visio are the *only* things I use windows for.
Simple stuff, like swapping channels, is very tedious in Audacity (I asked on the audacity mailing list and nobody had an easy method). Improving audacity all comes down to more 'man years' of work required.
Most Linux audio tools don't do 24 bit well. They quietly convert to 32 bit and back to 24, introducing noise. Ecasound, as good as it is, does not seem to be bit accurate. Kai does an amazing job of supporting ecasound but doesn't seem to 'get' the need for bit accuracy and that most hardware is 24 bits and not 32.
I'm sure that Linux audio will continue to improve. I was doing 24 bit laptop recording with linux for a while and it worked quite well (and I think ecasound was quietly converting the 24 bit data from my Minime to 32 bits and then back to 24 when writing it out...). Ultimately, I decided I wanted to make recordings more than I wanted to hack Linux sound and fight battles over bit correctness, etc. Sometimes you just want to drive the car, not build it