Basically we still need a VMFS3 datastore as a "AUX" to shrink disks.
This is interesting! Basically, logically, in plain English.. what happens is since the VAAI datamover used is not at the ESXi layer, the storage doesn't know what is on the VMDK and _has-to_ copy everything. There is no chance for ESXi layer to figure out which blocks to drop!
The conditions - ESXi 5.x onwards (VMFS5) + VAAI capable/enabled storage, Thin Provisioned VMs
"...When the source filesystem uses a different blocksize from the destination filesystem, the legacy datamover (FSDM) is used. When the blocksizes of source and destination are equal, the new datamover (FS3DM) is used. FS3DM decides if it will use VAAI or just the software component. In either case, null blocks are not reclaimed"
Thanks to Boon Hong for highlighting this.