Because it uses a randomized pivot, the worst case for Hoare's implementation has a probability on the order 1/n!. With twenty inputs, that's 1/(2.4 * 10^18).

Probabalistic analysis [1] is the difference between theory and practice. Hoare knew what he was doing and it's no accident was recognized with a Turing Award.

[1] https://www.cs.cmu.edu/afs/cs/academic/class/15451-s07/www/l...