Your idea actually only makes sense if load is not equally balanced. If it is equally balanced that means all servers will become "full" at approximately the sa

Your idea actually only makes sense if load is not equally balanced. If it is equally balanced that means all servers will become "full" at approximately the same time. Unless you exchange information about load between servers (which is difficult) this means a client will have to try a large number of servers until it finds one that is not "full".