That said with everything moving to the browser (or more correctly a zero install thickis client with remote storage) I am not sure it matter than much as more and more of the work seems to be done in browsers.
With asm.js as a potential LLVM target, I don't think it will matter. The browser could just serve as another compilation target with a specific foreign language interface.
Trying it out now ;)
I count "mainline Lisp", which ended with Common Lisp, and Scheme as the other two; others might count Autodesk's and Gnu Emacs' embedded Lisps.
Clojure is even showing signs of becoming mainstream in a way no Lisp has been since mainline Lisp through sometime in the '80s, not counting Scheme's niches in language research and (temporary? one) education. Or perhaps even better.
It is even possible that he viewed the JVM as a stopgap to be abstracted out when and if the language got popular. It seems they are moving in this direction with ClojureScript. Clearly he made this decision thoughtfully and consciously because he was an expert in C++ before he ever wrote a line of Clojure interpreter, and could have made it native if he wanted.
But the argument "Clojure is successful, therefore all his decisions must have been optimal" (caracaturing you a little bit) is obviously false. Replace "Clojure" with "Linux" or even "Windows" and you see the fallacy.
I still really enjoy Common Lisp (even though it seems like I spend my time with Clojure or Haskell) and Clasp, when it is mature, will be a good incentive to use Common Lisp for more development use cases. (It is easy enough to build stand alone executables with SBCL, CCL, and Franz, but having a smaller executable would be really good, if indeed the executables would be smaller with Clasp).