ReFS benefits with VHD

It does not write zeros to all extents of a pre-allocated file

Data can be merged between files through metadata operations instead of copying actual data (the metadata of the target is updated to point to existing clusters on disk currently used by the source disk)

Creating fixed size VHDs on ReFS is near instantaneous since it does not have to write zeros to all pre-allocated clusters which is the case with NTFS and instead uses metadata operations

Dynamic VHDs extend faster as once again the additional clusters added do not have to be zeroed which results in reduced latency of operations

Data is migrated from child VHDs to the parent VHD without data migration resulting in very fast merges including checkpoint clean-ups (and remember checkpoints are now used as part of the Hyper-V backup in Windows Server 2016)