Similar experience here. I wrote a log parser with indexing and hierarchical cache invalidation to deal with larger than memory amounts of logs quickly and in a natural coding style, and after that I was hooked. Tiny code base, good performance, little need for ugly hacks.
My latest site is now written in it and clojure script, and I see no reason to not continue. Though cljs is definitely a few years behind on the stability curve.
I haven't had any major issues with cljs yet. I'm using it for personal side-projects where I prototype random ideas along with Om and React.
I've had some issues with the Austin REPL however, there is sometimes random disconnections and it was hard to setup the first time, but using it is so much fun it completely shadows these issues.
I've switched to using the Weasel[1] REPL instead and haven't had any disconnect problems since. Its also a little easier to setup (although still not "easy").
My latest site is now written in it and clojure script, and I see no reason to not continue. Though cljs is definitely a few years behind on the stability curve.