You seem to be arbitrarily splitting on spaces. This will break unless there's a clear space either side of the +. A better option would be as follows.
my $text = '1+2<maths> + dfdf</maths>';
my $output = "";
while ($text =~ m/((<maths>.*?<\/maths>)|([^<]*))/gs) {
if ($2) {
$output .= $2;
} else {
my $segment = $3;
$segment =~ s/+/ + /g;
$output .= $segment;
}
}
print $output . "\n";
The basic premise being to scoop up and ignore (push onto output) anything in <maths>, or scope up as much that can easily be determined not to be in <maths> (ie no angle brackets) and parse that before putting it on the output.