Query plans are dependent upon actual data itself. Plan performance issues can be difficult to replicate in test envs.

AFAICT, this product relies on LVM snapshots/copy on write : so is ideal diagnosing, trying out ideas on large databases.