I'm wondering if one of the VMware people can answer this. I've read through the VDDK release notes, and see this as a carry over issue from VDDK 5.5 ... but never actually see it as a resolved issue.
Slow read performance with NBD transport.
Corresponding bugzilla number is 1376626.
I'm wondering in what version of VDDK/ESXi this may be fixed (if it is), and if there is a corresponding stack/backtrace to identifiy when no data is sent for some time.