Your "recursive subpattern" approach still needs an outer while loop, so I don't really see the point, apart of complicating it.

(I think /g stops if there is either no match or pos hits the end. To avoid an outer while-loop you might need to anchor a pattern at the start without changing pos , like inside a look-ahead )

FWIW: This would be my approach:

#  https://theweeklychallenge.org/blog/perl-weekly-challenge-340/
use v5.12;
use warnings;

for ('abbaca','azxxzy','aaaaaaaa','aabccba','abcddcba') {
  my $str = $_;
  say "*** Input: $str";
  while ($str =~ s/(\w)\1+//g) {
    say "Remove '$&' => '$str'"
  } ;
  say "Output: '$str'\n\n";
}
[download]

It's producing the same "Output" while skipping some steps.

(for exactly the same intermediate steps like in in the weekly challenge remove /g and +)

*** Input: abbaca
Remove 'bb' => 'aaca'
Remove 'aa' => 'ca'
Output: 'ca'


*** Input: azxxzy
Remove 'xx' => 'azzy'
Remove 'zz' => 'ay'
Output: 'ay'


*** Input: aaaaaaaa
Remove 'aaaaaaaa' => ''
Output: ''


*** Input: aabccba
Remove 'cc' => 'bba'
Remove 'bb' => 'a'
Output: 'a'


*** Input: abcddcba
Remove 'dd' => 'abccba'
Remove 'cc' => 'abba'
Remove 'bb' => 'aa'
Remove 'aa' => ''
Output: ''
[download]

Tho I can't guaranty that there isn't a case where only replacing 2 instead of 2 or more leads to different results.