The contents of comments are scanned, and therefore strings and
comments in comments, like
/* and one "nested" */,
need to be valid (properly closed). This enables commenting out a
random piece of contract that may already contain comments. The
downside is that you can only comment code or text whose strings and
comments are valid.
Comments can contain UTF-8 glyphs. Given that the generated Michelson does not contain comments, UTF-8 may matter for documentation of the source, and error messages display line offsets in the source as numbers of such glyphs, not bytes.