It's more a less that. In short, the new disk is created on both sides, and blocks mirrored on both sides.
First, the read only "parent" VDI with frozen blocks is copied, when it's done, the VM is briefly paused, the memory migrated to the destination and then unpaused on destination.
Consequently, only the blocks written during 11 hours will use space for new blocks on the source SR (and destination, since it's mirrored).