If I understand your question correctly (which is far from guaranteed), you might try something like:

#!perl -pi~
my $p;
my $done_replacement = 0;
while (<>) {
    if (/^A (.*) B$/) {
        $p = $1;
    }
    elsif (defined $p && !$done-replacement) {
        s/^(C .*)$p$/$1 Q/;
        $done_replacement = 1;
    }
}
[download]

Of course you must replace A, B, C and Q with your actual data, and pass the filename as a command-line argument to this script. It would be easier to give a precise answer if you could make your question more concrete by giving some example data. and saying what A, B and C really are.

I'm assuming that your A P B and C X P each constitutes an entire line of the file, because your use of “second line C X P” made me think that's what you meant.

Also, why is it important not to create an additional file? Usually it's safer to save the results of a transformation in a new file, in case something goes wrong. (In the code above, the switch -i~ causes the original file to be backed up with a name like oldname~, which gives you some protection at least.)