You are assuming binary computers can represent any decimal fraction exactly. They cannot (natively). Basically, what you are seeing are rounding errors.

Please read the FAQ for more details.