On my Macbook Pro, reading from standard input incurs a bunch of locking overhead, which complicates profiling and (IMO) adds noise to results. This adds the option to read from a file, which doesn't incur this overhead.
