Why not?

Here's one reason:

cmpthese -1, {
    a=>q[ 
        open O, '>', 'log'; 
        for(1..10000){
            print O chr(0)x80;
        }
        close O;
    ],
    b=>q[
        for(1..10000){
            open O, '>>','log';
            print O chr(0)x80;
            close O;
        }
    ] 
};;
            (warning: too few iterations for a reliable count)

     s/iter      b      a
b      1.30     --   -99%
a 8.42e-003 15300%     --
[download]

A second reason is that the first consumes almost no resource; whereas the second consumes prodigious amounts of it.

The times before and after the first:

a  5.546|5.343|0|0
a  5.546|5.343|0|0
[download]

And the second:

b  5.578|5.343|0|0
b  5.734|6.375|0|0
[download]

Whether you are just impacting every other process on your local server; or being billed for all that addition CPU & IO when running your process in the cloud; incurring 1000's of times more resource use in order to satisfy a dumb tool that is easily silenced is .... Its your dollar, you pick the adjective.