[ Update: The end of the post I replied to, starting at "It does have its limitations though" was not present when I replied (nor during the several times I had reason to load a page that included that node during the time I took composing my reply -- I don't believe it was even present immediately after I replied; though I am less sure of that). FYI. ]

Do you know that a reference to substr works as you described?

Only if you use a single substring.

They long since fixed the 'only one lvalue ref' limitation.

Thanks for the notice, but that isn't what I was talking about. Being able to have more than one lvalue substr ref doesn't itself result in "multiple simultaneous substrings of the same string such that they cooperate" nor to them "working as the OP described". (Nor does it lead to any particularly "tricky bits", certainly not any that qualify as "fun to hash out".)

#!/usr/bin/perl -w
use strict;

my $s = 'one,two,three,four';
my @p;
while(  $s =~ /([^,]+)/g  ) {
    push @p, \substr( $s, pos($s)-length($1), length($1) );
}

print "($$_)" for @p;  print " [$s]\n";

${$p[0]} = 'five';  # Change 'one' to 'five'
${$p[2]} = 'X';     # Change 'three' to 'X'

print "($$_)" for @p;  print " [$s]\n";

print $], $/;
__END__
(one)(two)(three)(four) [one,two,three,four]
(five)(,tw)(X)(r) [five,twoXe,four]
5.012000
[download]

While the OP was requesting an outcome like:

(five)(two)(X)(four) [five,two,X,four]
[download]

Note that I'm not trying to argue that \substr magic should be taught to "cooperate". What \substr magic does is quite simple and sane and I see no reason to change it. It just doesn't match what the OP was asking for (nor what my module implements).