Perhaps seeeing exactly what is being matched and captured will help (/x regex modifier used to help my poor eyes):

>perl -wMstrict -le
"my $s = qq{abandon/to\t(a road)\t};
 $s =~ s{ ([^\t]*) \t ([^\t]*) \1 ([^\t]*) \t }{\t$2$1$3\t}xms;
 print qq{1 '$1'  2 '$2'  3 '$3'};
 print qq{'$s'};
"
1 'o'  2 '(a r'  3 'ad)'
'abandon/t      (a road)        '
[download]

Update: This may help (note anchor at start of string): If the entire content of the first cell is repeated anywhere in the second cell, delete the entire content of the first cell:

>perl -wMstrict -le
"for my $s (qq{abandon/to\t(a road)\t},
            qq{abandon/to\txxx abandon/to yyy \t},
           ) {
   print qq{'$s'};
   (my $t = $s) =~ s{ \A ([^\t]+) (\t [^\t]* \1 [^\t]* \t) }{$2}xms;
   print qq{1 '$1'   2 '$2'} if defined($1) and defined($2);
   print qq{'$t' \n};
   }
"
'abandon/to     (a road)        '
'abandon/to     (a road)        '

'abandon/to     xxx abandon/to yyy      '
1 'abandon/to'   2 '    xxx abandon/to yyy      '
'       xxx abandon/to yyy      '
[download]

And this, capturing less, may be slightly better:
    s{ \A ([^\t]+) (?= \t [^\t]* \1 [^\t]* \t) }{}xms;