Microsoft has a seed finder specifically aimed at avoiding a priori bias in experiment groups, but IMO the main effect is pushing whales (which are possibly bots) into different groups until the bias evens out.

I find it hard to imagine obtaining much bias from a random hash seed in a large group of small-scale users, but I haven't looked at the problem closely.