The following code builds two 42 MByte test files (2 million lines each) then runs the analysis. The analysis phase takes about three minutes to run.

#!/usr/bin/perl
use strict;
use warnings;

my $kTestLines = 2000000;
my $kMinSame   = 2;
my $testA      = 'testA.txt';
my $testB      = 'testB.txt';

srand (1);
buildTestFile($_) for $testA, $testB;

open my $inA, '<', $testA or die "Can't open $testA: $!\n";
open my $inB, '<', $testB or die "Can't open $testB: $!\n";

my %aKeys;

print scalar localtime, "\n";

while (my $aLine = <$inA>) {
    chomp $aLine;

    my @keys = map {lc} split /\s+/, $aLine;

    push @{$aKeys{$_}}, $. for @keys;
}

while (my $bLine = <$inB>) {
    chomp $bLine;

    my @bWords = map {lc} split (/\s/, $bLine);
    my %lineAHits;

    ++$lineAHits{$_} for map {@{$aKeys{$_}}} grep {exists $aKeys{$_}} 
+@bWords;

    my @matchALines = grep {+$lineAHits{$_} >= $kMinSame} keys %lineAH
+its;

    next if !@matchALines;

    printf "%s:\n    %s\n", join (', ', @matchALines), $bLine;
}

print scalar localtime, "\n";


sub buildTestFile {
    my ($fName) = @_;

    open my $fOut, '>', $fName or die "Can't create '$fName': $!\n";

    for (1 .. $kTestLines) {
        my %words = map {$_ => undef} map {
            join '', map {randLetter()} 1 .. 4
        } 1 .. 4;

        print $fOut join (' ', keys %words), "\n";
    }
}


sub randLetter {
    return chr 65 + rand 26;
}
[download]

Prints (with about 2800 lines omitted):

Sat Feb 27 11:43:38 2016
704856:
    CYXG GWVB OYLX YNWJ
849378:
    ECML DIYS APPF OYLR
...
1090468:
    VSIR CKVJ GWIV IOXN
1327692:
    YJOQ YOJT NCZL VCSA
740815:
    ZZJN WYVG EETN QADD
Sat Feb 27 11:47:08 2016
[download]