In my understanding the reason lies in a way OS uses those file names. Unfortunately it always re-encodes filenames on the fly, at least for Russian encoding it switches between CP866, CP1251 and unicode

In my opinion this is mostly done to be compatible with C libraries, which are console-mode based.

When not in Russian but in any other encoding, including Far East, things more complicated.

Here is how I get away of the problem using OLE interface, with elder perl:

use Win32::OLE qw(in CP_UTF8);
use Win32::OLE::Const;

Win32::OLE->Option(CP=>CP_UTF8);

use Unicode::String qw/utf8/;

my $oshell = Win32::OLE->new('Shell.Application') or die "$@";
my $f = $oshell->NameSpace(Win32::GetCwd());
print "[$f]";
my $fi = $f->Items;
print $fi->Count;
print "\n";
for (0 .. $fi->Count-1) {
  my $item = $fi->Item($_);
  my $name = $item->Name;
  my $u=utf8($name);
  my $s = $u->hex;
  $s=~s/U\+00(\w\w)/my($r,$p)=((pack 'H*',$1),$&);if($r=~m(^[()\w .;\-
++!]$)){$r}else{$p}/eg;
  $s=~s/(U\+[\da-f][\da-f][\da-f][\da-f])/($1)/ig;
  my $ren=0;
  $ren=1 if $s=~/U\+(?!00)/;
  $s=~s/[ +]//g;
  print "$ren|$s\n";
  if($ren){$item->{Name}=$s}
}
[download]

It is quite possible for you to find another solution in Win32::xxxxxx modules.