Hardware fails. This architecture has single points of failures which means if one of those instances fails the entire system goes down. This is generally not acceptable for environments with 1 million active users (which implies high business value)