This is a different case. The original report used "qemu-img create" in step 2, which results in a sparse image that refers to the backing file for all data. Your sequence has "qemu-img convert" instead, which fully populates disk.qcow. Therefore, in step 3, "qemu-img convert" leaves the full allocation intact.
This is a different case. The original report used "qemu-img create" in step 2, which results in a sparse image that refers to the backing file for all data. Your sequence has "qemu-img convert" instead, which fully populates disk.qcow. Therefore, in step 3, "qemu-img convert" leaves the full allocation intact.