I suspect you are using a windows system, and Errto is on the right track: you need to take proper control of the "native crlf" behavior for that OS. Note that ordering of the PerlIO layers can be significant. I don't have a windows system to test on myself, but my bsd-based macosx shows the following behaviors with the various permutations - YMMV, but I think you'll see something like this:

$ perl -e 'open($fh,">:crlf:encoding(UTF-16BE)", "test.utf16"); print 
+$fh "hello\n"'

$ xxd test.utf16
0000000: 0068 0065 006c 006c 006f 000d 0a         .h.e.l.l.o...

# that was bad -- odd number of bytes


$ perl -e 'open($fh,">:encoding(UTF-16BE):crlf", "test.utf16"); print 
+$fh "hello\n"'

$ xxd test.utf16
0000000: 0068 0065 006c 006c 006f 000d 000a       .h.e.l.l.o....

# that was good.


$ perl -e 'open($fh,">:raw:encoding(UTF-16BE)", "test.utf16"); print $
+fh "hello\n"'

$ xxd test.utf16
0000000: 0068 0065 006c 006c 006f 000a            .h.e.l.l.o..

# also good (no CR, but who needs that anyway? ;)


$ perl -e 'open($fh,">:encoding(UTF-16BE):raw", "test.utf16"); print $
+fh "hello\n"'

$ xxd test.utf16
0000000: 6865 6c6c 6f0a                           hello.

# not what you want
[download]

Note that ":raw" is sort of a synonym for "unix" in this context (that is, the last two examples behave the same when using "unix" instead of "raw").

So, if you want a "standard" CRLF discipline for output, interacting correctly with UTF-16, there's only the one way to do that, it seems; OTOH, if you want unix-like LF discipline with UTF-16, there's a couple ways to get that (that is, you can say "unix" or "raw", but you still have to get the layers in the right order).

update: As almut wisely explains above, my examples are deficient -- each of the "working" cases should have the additional ":utf8" layer at the end, so that actual "wide characters" in the strings being printed will be interpreted and encoded correctly on output. Apologies for the confusion.