If we don't do this we end up using the generic read_exact method
which is not necessarily optimal. This is especially when
using a specialized Read implementation to go fast.
See #206