We have a GWT app with a bunch of Selenium acceptance tests. I thought (like you) that it would surely be safe to upgrade GWT from 1.7 to 2.0. And it was - mostly. The app still worked the same for "human" users, but the selenium tests all broke. There's a newer version of Selenium in preparation (at alpha release, with many UnsupportedOperations), but if we want to stay with GWT 2, it seems we have to give up some testability. So be careful about "future-proof" assumptions.
Our decision to use GWT was made months ago, after comparing YUI and ZK. I'm still glad we chose GWT. The level of support on the GWT website and the general quality of the documentation seems very high.
GWT does module splitting and provides performance profiling which help to counter arguments that it is not lightweight enough.
