I get what you're saying, but the human ear has it's limits. Blind A:B testing with CDs vs the higher-resolution SACDs has had mixed results - there are a very few people who can consistently tell the difference, but everyone else might as well be coin-flipping.

Yes. Plus the implementation filters can not be ideal as it would require infinite resources and in practical implementations have several species of non-idealities.