The toolbar buttons, for example are a great example of a major usability mistake: what we used to call "mystery meat navigation" back in the day, because it was difficult or even impossible to know in advance what would happen if you clicked a given button.
Forcing a user to stop and think about "what does this icon do" -- and yes, even your "heavy" users are going to run into this -- just to make something look pretty is basically always a bad idea.
No, this is incorrect. I'm a heavy user of Gmail, and I still can't find the damn new refresh button. How many weeks should it take for me to get used to this so-called "improvement". I preferred it when it was text. I just don't see the new spiral icon thingie as "refresh".
Another way to look at it is if you make an improvement to something I don't care about, it's not an improvement. Especially, if it makes my job harder.
You need to actually measure what users do, not generalize from unsupported assertions.