Why is it hashing files and not blocks? If a block is hashed and written there's no need to touch it again.
The real issue is that modifying large files on a COW filesystem causes it to spread that file's guts all over the disk (i.e. fragmentation), and results in serious performance loss over time.
For some reason that doesn't happen nearly as much on ZFS and it's totally fine for VM and database workloads.