In almost a decade of using Clojure, I've never seen persistent data structures become the bottleneck. The difference in cost is O(1) for in place mutation vs O(log 32 n) for revisioning. Note that the large exponent means the tree ends up being very flat, so updates end up being quite cheap in practice. Most times when I do something to a persistent data structure, I only change a small portion of the overall data structure. In fact, I think it's quite rare that you'd end up cloning the whole thing.
Personally, I view mutability as an optimization, and I think it should be used in cases where you actually need the performance beyond what persistent data structures can offer. My experience is that such cases are far and few between.