That doesn't look very deterministic to me! The following code may work a little more efficiently with large strings:

use strict;
use warnings;

my $incrsdel = 10;
my $str = "acbcybaycayacyccy";
my @chrs = split '', $str;
my $yTot = $str =~ tr/y//;

for my $letter ('a', 'c') {
    $yTot += subLetter ($letter, $incrsdel - $yTot, \@chrs);
    last if $yTot >= $incrsdel;
}

$str = join '', @chrs;
print "$str\n";


sub subLetter {
    my ($letter, $limit, $chrs) = @_;
    my @letPos = grep {$chrs[$_] eq $letter} 0 .. $#$chrs;
    my $count = 0;
    
    while ($count < $limit && @letPos) {
        $chrs->[$letPos[my $idx = rand @letPos]] = 'y';
        splice @letPos, $idx, 1;
        ++$count;
    }
    
    return $count;
}
[download]

Oh, and it scales to a larger set of replaceable characters easily too.

More possible solutions! I need a happy smiley.

Hi, Using RegExp I'd use the following (here @stash represents the layers):

use strict;
use warnings;

my $needed_ys = 7;
my $str       = 'Why do you need 7 ys when cll you hve re as and cs';
my $orig      = $str;
my @stash     = qw(a c);

my @existing = $str =~ /y/g;
my $missing = $needed_ys - @existing;

foreach my $letter (@stash) {

    while ( $missing > 0 ) {
       last unless $str =~ s/$letter/y/;
       $missing -= 1;
    }
}

print 'Original: ', $orig, "\n";
print 'Result:   ', $str, "\n";
[download]

You could check first whether there are likely to be enough a's and if not use tr/a/y/ to change them all and then go straight to the c's.

my $letter='a';
if (($cnd_a + $cnd_y) <= $incrsdel){
  $aob[$yb] =~ tr/a/y/;
  $cnd_y += $cnd_a;
  $letter='c';
}
while ($cnd_y < $incrsdel){
  my $xda = int rand( $chntot + 1 );
  if ( substr( $aod[$yb], $xda, 1 ) eq $letter ) {
     substr $aod[$yb], $xda, 1, 'y';
     $cnd_y++;
  }
}
[download]