Isn't it more bytes in practice? That's 88 bits, but they could be spread over different bytes, right?

In other words: shouldn't you calculate the loss over the total number of bits?