The following subroutine seems to return the same result as yours, and is about 2 times faster. You didn't provide much testing data, so I'm not sure it will work correctly for the real input.

sub not_so_special {
    my $s = shift;
    my @cols = split /;/, $s, -1;
    my $e;
    $cols[2] =~ s/=(.+)/'=' . (' ' x length($e = $1))/e;
    return join ';', @cols, $e
}
[download]

Update
Or, maybe, more readable and a tiny bit faster:

sub new {
    my @cols = split /;/, shift, -1;
    $cols[2] =~ s/=(.+)/=/;
    $cols[2] .= ' ' x length $1;
    return join ';', @cols, $1
}
[download]

Update 2
This seems even faster (but uglier):

use Syntax::Construct qw{ /r };
sub newer {
    shift =~ s/^([^;]*;[^;]*;)(.*?=)([^;]*)(;.*)/"$1$2" . (' ' x lengt
+h $3) . "$4;$3"/er
}
[download]