"Ultimately, isn't my saying "I like gnome3 better" the whole point here, just as (please correct me if I'm wrong) you're saying "I like gnome2 better"?"
While the field of HCI studies is not exactly comparable to exact sciences, a UI regression is something that can be measured by somewhat precise metrics (a feature was there and has been automagically removed because incompatible with the GNOME "brand", an action that required n mouse clicks now requires 2x n, an action that was perfectly visibile to the user is now hidden behind a modal screen, and so on).
So no, it's not that you can say "but I like it" and wave all problems away: your opinion is your opinion and can be shared or not, an UI regression is a UI regression and it's there to stay if nobody does something.