Have variable performance and emissions is not the problem. The EPA testing should (does?) test multiple types of usage and use an amalgamation of the results for an overall score.

The problem isn't that they are changing how they perform depending on circumstance, it's that they've defined a circumstance specifically for "we're being tested" and change how it functions based on that, which means the test no longer has any bearing on reality. That's the difference between a good feature and a defeat device.