Long running sub-query: bug or feature

Any idea why a query will use the correct index when run standalone; but will use an incorrect index when executed as a sub-query? We have a query similar to this:

select ...

   from table_a
   where col1 = 12345 and col2 in

      (select col2
          from table_a
          where col1 = 12345
          group by col2
          having sum(col3) != 0);

Table_a has an index on col1 and an index on col2. When the above sql statement is run, the execution plan for the sub-query incorrectly uses the index on col2. When the sub-query is run standalone, it correctly uses the index on col1. Is this a bug or a feature? We are running rdbms 7.1.6 and use the rule based optimizer.

-- 
Kelly Young
Database Administrator
Maricopa Community Colleges
young_at_maricopa.edu