When working with real-world statistics? Absolutely, there is no reason to believe that this assumption holds.

But the context is randomized (quantum) algorithms, where this is trivial to guarantee.