This is the modified code:

use strict;
use warnings;

use Benchmark qw( cmpthese );

my @strings = qw(
  USERID1|2215|Jones|
  USERID1|1000|Jones|
  USERID3|1495|Dole|
  USERID2|2500|Francis|
  USERID2|1500|Francis|
);

cmpthese(
    -1,
    {
        outside => sub {
            my ( $x, $y, $z );
            for (@strings) {
                ( $x, $y, $z ) = split /\|/;
            }
        },
        outside2 => sub {
            my ( $x, $y, $z );
            for (@strings) {
                ( $x, $y, $z ) = split /\|/, $_, 3;
            }
        },
        inside => sub {
            for (@strings) {
                my ( $x, $y, $z ) = split /\|/;
            }
        },
        inside2 => sub {
            for (@strings) {
                my ( $x, $y, $z ) = split /\|/, $_, 3;
            }
        },
    }
);
[download]

$ perl bench_inside_outside.pl
             Rate  outside outside2   inside  inside2
outside   90269/s       --     -20%     -40%     -51%
outside2 113390/s      26%       --     -25%     -39%
inside   151060/s      67%      33%       --     -19%
inside2  185735/s     106%      64%      23%       --
[download]

That second point is interesting, because I have experienced in the past that specifying a limit brings an improvement when the string being split would yield (without limit) more fields than the limit, presumably because Perl is able to stop processing the string as soon as the limit is reached, but I would have thought that this advantage would to a large extent vanish when the limit is the same as the number of potential fields in the string being split. Good to know. Thank you choroba for this comment.