I think Word is probably half-responsible for the mangling here. If it's trying to display each byte, then it means it's not actually reading it as UTF-8, but in some other encoding!

I'll give an example using Windows. First, here's utf8.pl:

# U+73E0 ("pearl")
print "\xe7\x8f\xa0";
[download]

Now, I execute that and redirect it to both utf8.html and utf8.txt.

Chrome displays the character correctly, because it assumes UTF-8 by default. Notepad also appears smart enough to guess the encoding.

On my system at least, opening the file with Word prompts me to select the encoding; and by default, it guesses UTF-8 and renders the character correctly. Note that if I pick "Windows (Default)" or "MS-DOS", I get garbage.

So try messing with Word a bit; if you use the File -> Open menu (instead of just opening the file from Explorer directly), you can get additional conversion options (sometimes!).

Anne