GNU yes(1) is not that fast!
This implementaion is not only about 8
times as fast[0], it uses half as much CPU.

Note that this implementation is not even
optimal, the overhead at the start can be
reduced by creating a buffer as large as
the pipe and fill that buffer and only do
one write(2) or vmsplice(2) to the pipe.
Speaking of this overhead, this implementation
is completely useless[1] unless the other
program is will even read {PIPE_BUF} bytes
(4096 on Linux, 512 on POSIX) + 2²⁰−2¹⁶ bytes
(on Linux, unspecified for POSIX). Therefore,
this implementation of yes(1) is just silly
and should not be used by anyone.


[0] On my computer. If you get different
    results please leave a comment.

[1] Has no benefits what so every in any
    aspect at all.