Improving Fortran Results in the Julia Micro-benchmarks

I feel this changes the logic of the benchmark. I guess the goal of that microbenchmark is to test the system calls. The above code seems only calling write once.