There is a gotcha with the solution as written (well, several actually) - you need to provide every possible (or at least likely) case variation for each of the matched words. Combining the special markup you have alluded to elsewhere with the e flag for the regex gives a somewhat more robust solution:

use strict;
use warnings;

my $text = join '', <DATA>;
my %replacements = (
    'romeo'  => 'CountZero',
    'juliet' => 'Elisabeth',
    'rom.'   => 'CountZero',
    'jul.'   => 'Elisabeth',
    'abram'  => 'Igor',
    'abr.'   => 'Igor',
);
$text =~s/%%([^%]+)%%/replace($1)/ge;

print $text;


sub replace {
    my ($needle) = @_;

    return "%%$needle%%" if ! exists $replacements{lc $needle};
    return $replacements{lc $needle};
}


__DATA__
(follows the full Project Gutenberg e-text of Shakespeare's %%Romeo%% 
+and %%Juliet%%;
found at http://www.gutenberg.org/dirs/etext97/1ws1610.txt)
[download]

Prints:

(follows the full Project Gutenberg e-text of Shakespeare's CountZero 
+and Elisabeth;
found at http://www.gutenberg.org/dirs/etext97/1ws1610.txt)
[download]