How about:

use if $] ge '5.016', feature => 'fc';
use if $] lt '5.016', 'Unicode::CaseFold' => 'fc';
[download]

See Unicode::CaseFold. Otherwise you could do:

*case_insens = $] ge '5.016'
    ? sub { &CORE::fc($a) cmp &CORE::fc($b) }
    : sub { lc $a cmp lc $b };

@list = sort {case_insens()} @list;
[download]

Update: Used &CORE::fc() instead of CORE::fc() so that it really does work on older Perls.

Great idea and nice implementation, but I wonder whether it wouldn't be simpler at this point to use the case_fold function of Unicode::CaseFold in both cases, i.e. irrespective of the Perl version. Something like this (not able to test right now):

use Unicode::CaseFold qw(case_fold !fc)
@list = sort {case_fold($_)} @list;
[download]

(I know this no longer answers the OP exact question, but it seems to me this would be more straight forward.)

simpler at this point to use the case_fold function of Unicode::CaseFold in both cases

Sure, TIMTOWTDI :-) Personally I just prefer core Perl when reasonably possible.

Here's another variant (Unicode::Collate has been in the core since 5.8), although the sorting order is different:

use Unicode::Collate;
@list = Unicode::Collate->new()->sort(@list);
[download]

I've been mulling this over for a bit and expected this to show equality:

$ ./1.version.pl 
a is 5.026001
b is v5.26.1
version is v5
version is v5.26.1
c is ж
d is Ж
did not match
$ cat 1.version.pl 
#!/usr/bin/perl -w
use 5.011;
use Path::Tiny;
use POSIX qw(strftime);
use if $] ge '5.016', feature => 'fc';
use if $] lt '5.016', 'Unicode::CaseFold' => 'fc';

my $a = $];
say "a is $a";

my $b = $^V;
say "b is $b";

printf "version is v%d\n", $a;  # Perl's version
printf "version is v%vd\n", $b;  # Perl's version

my $c = "ж";
say "c is $c";

my $d = "Ж";
say "d is $d";

if (fc($c) eq fc($d)){
say "expressions matched";

} else {
    say "did not match";
}
__END__ 
$
[download]

The characters are ж and Ж .

my $c = "ж";

That's an HTML entity and consists of 7 separate characters. If you typed it like this then hopefully it becomes obvious why it doesn't match against "Ж" which is an entirely different string, folding notwithstanding.

OTOH, if you didn't type it like this but instead used a unicode literal string such as

my $c = "ж";

then your code is missing the absolutely necessary use utf8; pragma in order to decode this character properly.

(Edited for typo)

Isn't this easier?

use if $] ge '5.016', feature => 'fc';
BEGIN { eval 'fc 1' or eval 'sub fc ($) { lc $_[0] }' };

...;

@list = sort { fc $a cmp fc $b } @list;
[download]

This should be a Babble plugin!
map{substr$_->[0],$_->[1]||0,1}[\*||{},3],[[]],[ref qr-1,-,-1],[{}],[sub{}^*ARGV,3]

When I ran into this I came up with

BEGIN {
    *fold_case = CORE->can( 'fc' ) || CORE->can( 'lc' ) ||
        sub { lc $_[0] };
}
[download]

The explicit sub is because if you go back far enough lc() is not exposed in the CORE:: name space.