What CoffeeScript did introduce people to was the idea of a language that compiled to only Javascript. And as such, it didn't have a pre-existing language community to point out the shortcomings of the JS implementation. The problem with most compile-to-JS languages is that you never get the semantics exactly right; things like float semantics, string encodings, prototype chains etc. usually make the JS version just different enough to hold surprises for programmers used to the server-side implementation, which defeats the purpose of using the same language for both (Rhino had the same problem, which is why it took V8 before Node could really catch on). CoffeeScripters have no such preconceptions, though - it's a completely new language, so it's adopted on its own merits as a browser-side language by those who like what it provides.
I would actually disagree. In the Clojure world, we try to embrace the platform, meaning that we do not try to make all source compatible.
The goal is to have the basic comparable and then make all the more algorithmic library work everywhere. I think this has worked out pretty well so far.
Things like core.logic, core.match, core.typed and so on work on both sides with relatively small adjustments.
