Use compiled regular expressions as the hash keys. This stops you from rebuilding the regex on every iteration. Alternatively, memoize buildRegexp. Or Tie::RegexpHash might be useful.

You can eliminate a for loop and the long ||'ed regexp by changing the data structure. Don't group the subset names in an array to be used as the key. Put each subset name in the hash as a single key but have the value of the hash as a reference to the array. Equivalent subset names point to the same array.

my %lookingFor;
my $ref = [1, 2, 3];
$lookingFor{qr/a.*blah/o} = $ref;
$lookingFor{qr/the.*blee/o} = $ref;

#Tie::RegexpHash would let you do $lookingFor{$bigfileline}
#without looping over keys() at all.  No idea if it's faster.
[download]

The bottleneck is probably IO, but if it's not a more sophisticated data structure might help. You're not using the hash as a hash, but really as an array of arrays, i.e. an associative list. Is %lookingFor really big? If it is you might figure out a hierarchy and make it into a tree so you don't have to search linearly.