The redundancy of having five servers isn't helping if I'm always served by server #3, whereas an alternate method where any one of the five can answer my current request allows a machine to fail, or be taken down for upgrading, and the load on the other four just increases by 25% each.