While lisps do traditionally provide repl-driven development and you can do this in Clojure, most Clojure devs I know are not actively working right at the repl. If you do like some aspects of repl-like interactivity, Elm does have elm-reactor which has some similarities to Clojurescript's figwheel, though not quite as mature or comprehensive.

Aside from the repl, you are right that you can get a proof-of-concept going in Clojurescript extremely fast so if you want to see if an idea succeeds or fails before investing lots of time, you can't beat Clojurescript.

The Elm toolchain is slowly evolving to include things like dead code elimination (which it doesn't have yet but probably will in the near future). However, I think it will be a long time (if ever) that Elm will support global inlining, automatic variable renaming and some other optimizations that come for free with the Google Closure compiler support. The other important difference is that as the Google Closure compiler adds features or further improves its optimizations, these are mature and widely used and immediately available for free without much (if any) extra work on the Clojurescript team. It's a philosophical difference that the Elm devs prefer to solve these problems on their own rather than leverage existing tools in the industry, but since the Elm dev community is small (much of it is only one person), this means that lots of time must be spent on these solutions at the expense of features more specific to the Elm language.

Also I think you are right that Elm as a language is a little easier to learn than Clojurescript (though to someone new to FP, both Elm/Haskell style syntax and Lisp syntax might be equally bizarre). But I think that ease is made up for by some of the concepts in Elm that can be difficult to learn (decoding, Tasks, ports, etc), so it's probably a net flat difference on the learning curves between the two systems.