Dynamic scope in Clojure is a key way to model supervised local side effects (side reads and side mutations) in a tightly controlled way while ensuring that the expression as a whole remains referentially transparent and the effect behaviors are late bound. If you are into crazy metaprogramming, this is pretty important imo. The only other way to model effects that I know of is monads which involve lots of opaque closures, which block metaprogramming. Code-as-data captures a lot of static information about your program which monads lose.
I have very little interest in the ML family (well, let's just say that my interest in ML is in a different part of my brain that I don't want interfering with my thinking/exploration about systems like lisp/scheme and what kind of metaprogramming can be done with them).
I guess I'll have to think about dynamic-scoping a little more.
I mean once you have a function object instance, you've lost the source code, you can't know anything about what it does (beyond the type) except what you can glean by running it.
