It avoids dividing by 8 when deserializing a tx, which is a slow operation, and multiplies by 8 when verifying and extracing the amount, which is much faster as well as less frequent