This reduces the overhead of allocating, zeroing and GC-ing large
amounts of memory.

For large reads into /dev/null, this provides a 22% speed-up.