This is effectively how the split and shrink feature works.
There's a variety of features in Elasticsearch that help achieve the same kinds of elasticity goals. For instance, if you add nodes to your cluster then the shards will spread out so that each one has access to more resources (CPU, memory, etc), and you can shrink the cluster back down again when the extra power is no longer needed. You can also dynamically adjust the number of replicas in each index to cope with changing search load. Rollover is a good way to adapt to widely varying indexing load, and ILM helps to integrate this with shrinking and force-merging and other good practices. Elasticsearch also supports heterogeneous clusters allowing you to choose cheaper hardware for the nodes that don't need as much power.