[blippage]

As an experiment I tried

yes | pv > /dev/null

and was getting about 5.4GiB/s

On the fasm code, I was getting a meagre 7.3MiB/s. Ouch! The non-assembly version is considerably faster. I wonder if it is because I make a syscall for every write I want to perform, whereas C uses buffering, or something.