If you restore Thick Lazy Zero disk using SAN mode, overall restore performance can be affected because of round trips through the disk manager APIs, AllocateBlock and ClearLazyZero requests. The reason that NBD mode can work faster in comparison to SAN mode (when not using block-sized writes) is because the write is performed by the host which has inside knowledge about layout of the vmdk. This inside knowledge must be requested from the host when VDDK is doing direct writes to the LUN.