I think the thing here is that in, say, Scheme, the performance and actual execution semantics of your function vary wildly depending on how much you mangle you

I think the thing here is that in, say, Scheme, the performance and actual execution semantics of your function vary wildly depending on how much you mangle your algorithm. Precisely the same thing is true with recur in Clojure. The only difference is that once you've mangled your algorithm, you get to proudly signpost it by wrapping it in a recur. It doesn't help the machine at all, it just helps the user. Sounds like a job for... comments! (More specifically, recur only works in situations where TCO would have happened automatically in Scheme---and not even all of those, as other commentors have pointed out. So what you get is a weakened form of TCO that only kind of works and makes your code more verbose for no reason: as I said above, a comment would suffice in these circumstances.)