Think about where you would use CLJS - my use cases would not have an issue with extra 100kb of code - you don't use CLJS to do jQuery animations on your web page - you would use it to build complex single page apps or write server side code/scripts.
You don't use clojure for performance anyway, it's going to be slower by default (because of immutability/persistent data structures, and yeah I know about react benefits with immutability that's not my point - you're still going trough a lot more memory and stressing GC) - you use it to help you deal with your code because of it's semantics.
But in reality last time I tried CLJS I didn't really feel like it delivers on the productivity part and it's mostly because of implementation issues. IMO the decision to implement CLJS on top of Closure compiler and in Clojure (instead of going for a self hosting compiler) was a mistake - you can't overstate the value of REPL and fast iteration in a language like Clojure - and my last attempt to use CLJS the compiler/REPL environment far from what I would consider fast iteration : compiler took forever to start because of JVM and while it could run as a service it needed to be restarted frequently enough that it mattered, REPL was very unstable it would just die randomly - sometimes you'd need to refresh the page, sometimes you'd need to restart the server process. Oh and don't even get me started on the voodoo needed to get the damn thing running - install piggieback, then install austin and then add this weasel thingy then configure the server process all so you can get a halfworking repl and pray it doesn't break because good luck figuring out what's actually going on. Compare this to JS where I just go in to the devtools panel and test my code.