> What does non-8-bit mean? The standard demands >= 8 bits per char.

Platforms which do not allow you to address octets of memory.

ALUs and arithmetic having exotic "natural sizes" like 24 bits are not
a problem and can be different from the byte size. As you've said,
that can be covered by intrinsics and whatnot.

The smallest addressable size is more limiting because you have to be
able to modify bytes without data races/any impact on adjacent bytes,
and that's impossible or costly to emulate. If you can only modify 16
bits at a time, how do you write one octet of memory without a data
race, strictly guaranteed?

> It does not make sense to support the latest C++, but then use an exotic byte size.

Agreed.