Okay. So the optimiser favours the lookaheads by a factor of 16x:

#! perl -slw
use strict;
use Benchmark qw[ cmpthese ];
use List::Util qw[ shuffle ];

our @terms = qw[
    the quick brown fox jumps over the lazy dog
];

our $re = join'', map "(?=^.*$_)", @terms;
$re = qr/$re/;

our @lines = map join( ' ', shuffle @terms), 1 .. 100;
push @lines, ( 'every good boy deserves food' ) x 100;

our( $a, $b ) = (0) x 2;
cmpthese -1, {
    a=>q[ /$re/ and ++$a for @lines; ],
    b=>q[ for my $str ( @lines ) { !grep( $str !~ /$_/, @terms) and ++
+$b; } ],
};

print "$a:$b";

__END__
C:\test>junk48
    Rate     b     a
b 81.5/s    --  -94%
a 1399/s 1616%    --
208200:12000
[download]

Or, if I don't pre-compile the regex, 30x faster:

#! perl -slw
use strict;
use Benchmark qw[ cmpthese ];
use List::Util qw[ shuffle ];

our @terms = qw[
    the quick brown fox jumps over the lazy dog
];

our $re = join'', map "(?=^.*$_)", @terms;
#$re = qr/$re/;

our @lines = map join( ' ', shuffle @terms), 1 .. 100;
push @lines, ( 'every good boy deserves food' ) x 100;

our( $a, $b ) = (0) x 2;
cmpthese -1, {
    a=>q[ /$re/ && ++$a for @lines; ],
    b=>q[ for my $str ( @lines ) { !grep( $str !~ /$_/, @terms) && ++$
+b; } ],
};

print "$a:$b";

__END__
C:\test>junk48
    Rate     b     a
b 82.7/s    --  -97%
a 2632/s 3082%    --
389700:12000
[download]

It's faster, but not because the pattern itself is faster. You're getting the benefit because your regexp isn't changing, and you're doing the same pattern tens of thousands of times. The loop I suggested doesn't have that benefit.

But it can.

use strict;
use warnings;
use 5.010;
use Benchmark qw[ cmpthese ];
use List::Util qw[ shuffle ];

our @terms = qw[
    the quick brown fox jumps over the lazy dog
];

our $re = join'', map "(?=^.*$_)", @terms;
$re = qr/$re/;

our @lines = map join( ' ', shuffle @terms), 1 .. 100;
push @lines, ( 'every good boy deserves food' ) x 100;
my $line = join '&&', map {"/$_/"} @terms;

our( $a, $b, $c ) = (0) x 3;
cmpthese -1, {
    a=>q[ /$re/ and ++$a for @lines; ],
    b=>q[ for my $str ( @lines ) { !grep( $str !~ /$_/, @terms) and ++
+$b; } ],
    c=>qq [$line && ++\$c for \@lines;],
};

say "$a:$b:$c";
__END__
    Rate     b     a     c
b 39.8/s    --  -95%  -97%
a  807/s 1927%    --  -36%
c 1254/s 3051%   55%    --
109400:4800:243000
[download]

Ah well. If we're allowed to cheat then the difference moves into the realms of run to run variability :)

use strict;
use warnings;
use 5.010;
use Benchmark qw[ cmpthese ];
use List::Util qw[ shuffle ];

our @terms = qw[
    the quick brown fox jumps over the lazy dog
];

our $re = join'', map "(?=^.*$_)", @terms;
#$re = qr/$re/;

our @lines = map join( ' ', shuffle @terms), 1 .. 100;
push @lines, ( 'every good boy deserves food' ) x 100;
my $line = join '&&', map {"/$_/"} @terms;

our( $a, $b, $c, $d) = (0) x 4;
cmpthese -1, {
    a=>q[ /$re/ and ++$a for @lines; ],
    d=>qq[ /$re/ and ++\$d for \@lines; ],
    b=>q[ for my $str ( @lines ) { !grep( $str !~ /$_/, @terms) and ++
+$b; } ],
    c=>qq [$line && ++\$c for \@lines;],
};

say "$a:$b:$c:$d";

__END__
C:\test>junk48
    Rate     b     a     d     c
b 82.7/s    --  -97%  -97%  -99%
a 2657/s 3112%    --   -3%  -56%
d 2748/s 3222%    3%    --  -54%
c 5982/s 7132%  125%  118%    --
388200:12000:829800:389700
[download]

---

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

Touché :)++