$| = 1;
for ( my $len = 1; ; ++ $len ) {
    my $rx = rx1( $len );
    my @found = /$rx/ or last;
    print "$len: " . join( '', @found ) . "\n";
}

sub rx1 {
    # Cheap but wasteful.
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) C
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) CP
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) CPM
    #  0 wallclock secs ( 0.01 usr +  0.00 sys =  0.01 CPU) CPME
    #  1 wallclock secs ( 0.61 usr +  0.00 sys =  0.61 CPU) CS201
    #  4 wallclock secs ( 4.24 usr +  0.00 sys =  4.24 CPU) CS201G
    # 24 wallclock secs (23.48 usr +  0.01 sys = 23.49 CPU) CS201GA
    # 108 wallclock secs (105.62 usr +  0.02 sys = 105.64 CPU) CS201GAM
    # 402 wallclock secs (393.05 usr +  0.05 sys = 393.10 CPU) CS201GAME
    # 3917 wallclock secs (3827.03 usr +  0.50 sys = 3827.53 CPU)
    # 
    # real    74m16.003s
    # user    72m34.080s
    # sys     0m0.588s
    my $len = shift @_;
    my $pat = "\\A[^\n]*?" . ( "(?:(\\S)[^\n]*?)" x $len ) . "\n"
        . "(?:[^\n]*?" . join( '', map "\\$_\[^\n]*?", 1 .. $len ) . "\n)+\\z";
    return qr/$pat/;	
}

sub rx2 {
    # Expensive but less wasteful.
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) C
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) CP
    #  0 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU) CPM
    #  0 wallclock secs ( 0.03 usr +  0.00 sys =  0.03 CPU) CPME
    #  2 wallclock secs ( 1.66 usr +  0.00 sys =  1.66 CPU) CS201
    # 13 wallclock secs (13.27 usr +  0.00 sys = 13.27 CPU) CS201G
    # 83 wallclock secs (80.97 usr +  0.00 sys = 80.97 CPU) CS201GA
    # 413 wallclock secs (403.82 usr +  0.04 sys = 403.86 CPU) CS201GAM
    # ... to be continued (it's still working)
    # ... nope. It's 4x slower than the prior. I'll not bother.

    my $len = shift @_;
    my $pat = "\\A[^\n]*?" . ( "(?:(\\w)[^\n]*?)" x $len ) . "\n"
        . "(??{ qq<(?:" . join('', map { "[^\n>.\$$_.qq<]*>.\$$_.qq<" } 1 .. $len) . "[^\n]*\n)+>})\\z";
    use re 'eval';
    return qr/$pat/;	
}