What overhead should BHyVe incur vs. guest directly on bare metal, for memory mapped and non-memory-mapped IO to block device within the guest?

[Tinker]

Just to get your rough approximation.

OpenBSD guest.

E5 Xeon CPU. SSD:s with 4 or 16KB physical sector size.

In the virtualized case, the guest accesses the physical storage via a 
virtio block device, and the host keeps the image within a ZFS (so it 
gets ZFS' checksumming) and