There is an attribute called NOCOW that can be set on specific files that should not be copy-on-write, which is what messes with databases, filesystem images and other things that needs fast in-place updates.

It can also be a set as a flag on subvolumes.

I think ZFS was just ahead. We considered btrfs many years ago to serve up VM's for ESX via NFS, but it just wasn't as performant unless you ran in async mode which jeopardized data integrity. ZFS let you introduce SSD-based ZIL and L2ARC caching which made performance totally fine in sync mode.

We're mostly NetApp AFF these days, but early on had close to a petabyte of ZFS-based storage power VM's on SuperMicro or Dell gear. Definitely was higher touch than NetApp but far less expensive.

ZFS on Solaris 10 was ill suited for almost everything out of the box and was not feasible for MongoDB due to the half are integration between ZFS and the rest of Solaris