Who is that "we" that you're talking about?
EDIT: Well ok, JMC wasn't sure so I'll just say this more clearly: It's very likely that the previous code was just completely broken and worked by accident. We have a floating point test suite and Dolphin currently fails 20-30% of the tests (even though the test suite really is not all that extensive). So basically to exaggerate a bit to show the point, we now have 95% broken code instead of 99%, and some games apparently want us to be either completely accurate or completely broken, and they'll get confused otherwise.
EDIT: Well ok, JMC wasn't sure so I'll just say this more clearly: It's very likely that the previous code was just completely broken and worked by accident. We have a floating point test suite and Dolphin currently fails 20-30% of the tests (even though the test suite really is not all that extensive). So basically to exaggerate a bit to show the point, we now have 95% broken code instead of 99%, and some games apparently want us to be either completely accurate or completely broken, and they'll get confused otherwise.