IBM recommends no less than 10 shards per container. If we have less than 10 shards per container, the difference in load starts on a noticeable increase. Hopefully, this situation happens only at or near the end of the planned grid lifetime. If each container holds just one shard, then we have reached the end of the road for scalability. Should you need to grow the grid beyond the planned number of JVMs, an outage is unavoidable.
Avoid this situation with realistic planning up front to determine how many JVMs are required. Despite good planning, we can always reach the point where we have too few shards per JVM. If the planning stage used accurate data for determining the number of JVMs required, and shards/JVM is dropping, then you're experiencing faster-than-expected growth.
If you reach the point where five years of expected resources are simply not enough, then you have an enviable problem. Popularity to the point of burning through five years of data grid capacity is...