You might be able to use the (??{ ... }) construct to construct a regex with each line anchored at the right place. I haven't worked through the details, but it seems possible. Multi-dimensional pattern matching is actually an area of active research, and there are some interesting (and complex) algorithms out there to do it. Depending on time and level of interest, you might want to take a look on citeseer for "two dimensional pattern matching".

/s
Update: So, I actually tried this, and found you probably don't want to go this way:

my $pat = qr{^
         (?{$pre0 = pos})
         (.*)
         (?{$x = pos() - $pre0; $pre = qr/.{$x}/; print STDERR "$pre\n
+"})
         aa
         (?{$post0 = pos})
         (.*)
         (?{$x =pos() - $post0; $post = qr/.{$x}/; print STDERR "$post
+\n"})
         $
         ^(??{ $pre }) bb (??{ $post }) $
         ^(??{ $pre }) cc (??{ $post }) $
}xm;
[download]

is intended to match

aa
bb
cc

but in fact yields:

perl% perl -w 2dpm.pl
(?-xism:.{8})
zsh: 4283 segmentation fault (core dumped)
perl% perl -v

This is perl, v5.6.1 built for ppc-linux-thread-multi

Not sure if this is my fault or not.