I don't think you can recommend Quicksort unless you know something about the distribution of the elements to be sorted, or am I mistaken. After all, Quicksort can be very slow under certain conditions.
Quicksort's worst case involves bad pivot picking. The randomized version pretty much makes that a non-issue unless you're incredibly unlucky. Even still, with median-of-medians you can guarantee a deterministic worst case of nlog(n), so that's not true.
You can keep track of the performance of Quicksort then fallback to Insertion Sort if Quicksort is sucking. This is how std::sort() works in most C++ implementations.
