Using match time pattern interpolation can avoid the && and it also avoids false positives with overlapping words. I'm not sure you could call it better though. What you have is much easier to understand and maintain.

use strict;
use warnings;

use re q{eval};

my %alt = ( this => q{that}, that => q{this} );
my $re  = do   
   {
       local $" = q{|};
       qr{(?x) ( @{ [ keys %alt ] } ) .* (??{ $alt{ $1 } }) };
   };
print
   sprintf( q{%-9s: }, $_ ),
   m{$re}
   ? qq{matched\n}
   : qq{did not match\n}
   for qw{ thisthat thatthis thathis thisnthat thatnthis };
[download]

The output.

thisthat : matched
thatthis : matched
thathis  : did not match
thisnthat: matched
thatnthis: matched
[download]

I hope this is of interest.

Update: I noticed the i flag on one of the OP's patterns but the above didn't work with mixed case. It is fixed in this version.

use strict;
use warnings;

use re q{eval};

my %alt = ( this => q{that}, that => q{this} );
my $re  = do   
   {
       local $" = q{|};
       qr{(?xi)
           ( @{ [ keys %alt ] } )
           .*
           (??{ qr{(?xi) $alt{ lc $1 } } })
       };
   };
print
   sprintf( q{%-9s: }, $_ ),
   m{$re}
   ? qq{matched\n}
   : qq{did not match\n}
   for qw{ thisthat thatthis thathis thisnthat thatnthis ThISnthAt };
[download]

Which produces

thisthat : matched
thatthis : matched
thathis  : did not match
thisnthat: matched
thatnthis: matched
ThISnthAt: matched
[download]