I'm not sure that it is a different question? Anyhow, my limited testing threw up one case where the negative lookbehind was necessary:

With and without negative lookbehind:

Perl> 'aaabb' =~ m[ 
    (?{ local( $a, $b ) }) 
    ( 
        (?<!a) 
        ( a (?{++$a}) )+ 
        ( b (?{++$b}) )+
    )
    (?(?{ $a+1 == $b }) (?=) | (?!) ) 
]x and print "'$1'";;

Perl> 'aaabb' =~ m[ 
    (?{ local( $a, $b ) }) 
    ( 
        ( a (?{++$a}) )+ 
        ( b (?{++$b}) )+ 
    )
    (?(?{ $a+1 == $b }) (?=) | (?!) ) 
]x and print "'$1'";;
'abb'
[download]

A breif further test suggests that the negative lookahead may be unnecessary, but I threw it in on the basis that it would do no harm, and may catch a case I haven't thought to test. With a regex using closures, it's doubtful if it will slow things down noticably?

I meant I thought I was reading the OP's earlier thread, Extracting the number of repetitions from a regex. Sorry for the confusion.