regressions due to "users are used to" usually goes away after a while, you'll see a temporary dip in the results.

If you are convinced it's better for the long tun then what's the point in measuring ?