[masklinn]

> It appears that GNU yes will not do that and simply does unaligned writes in this case, which is likely to be considerably slower (possibly slower than write would have been).

Why would it be slower to do a single, say, 8190 bytes write instead of 2730 3-byte writes?

[AstralStorm]

Generally writes not aligned to cache line are slightly slower on most common architectures, vastly slower on others. (Such as many MIPS)

Small write calls themselves incur a considerable syscall overhead.