I may be missing the point here (I'm not a windows user either), but in the code snippet you showed in your "3rd update", I didn't see any evidence that the file handle was getting a utf8 IO layer (or alternatively, that the data being read in was getting "decoded" into perl-internal utf8).

Did you try doing the open like this?

open( READ, "<:utf8", "test.txt" );
[download]

You should also be setting the utf8 IO layer on your output file handle too:

binmode STDOUT, ":utf8";
[download]

I gather that your "resulting text" should have been different from the input text, but wasn't. This would be expected if your regex substitutions are based on unicode classes, but perl doesn't know that the strings are perl-internal utf8 unicode.