Postgresql results

RETURN QUERY EXECUTE was introduced with Postgres 8.4.
Your version is just too old and unsupported by now. Upgrade to a more recent version.

Also, dynamic column names in the result are very hard to come by. It's a principle of SQL that it wants to know the return type - including the names - up front.

Returning anonymous records without a column definition list only works for a single record. Else you have to provide a column definition list in the call. Details under his question on SO:
Return multiple fields as a record in PostgreSQL with PL/pgSQL

There are limited ways around this with with polymorphic types. Advanced stuff:
Refactor a PL/pgSQL function to return the output of various SELECT queries

BTW, your function would look like this in modern PL/pgSQL:

CREATE OR REPLACE FUNCTION get_policy_name (
  _id        int,
  _lang      text,
  _def_value text
) RETURNS TABLE (col text) AS
$func$
BEGIN
   RETURN QUERY EXECUTE
   format('SELECT COALESCE(%I, col, $1)
           FROM my_table WHERE id = $2'
         , 'col_' || _lang)
   USING _def_value, _id;
END
$func$ LANGUAGE plpgsql;

Call:

SELECT col AS col_fr FROM get_policy_name (1, 'fr', 'foo');

Here, I am simply using a column alias in the call to achieve what you want. Much easier than dynamic column names ...

It's a textbook case for a CTE, like @Daniel commented.
The example can be simplified some more. And you need to be aware of how LIMIT works in a UNION query.

CREATE OR REPLACE FUNCTION featured_products(valid_to_in      timestamp
                                           , taxonomy_id_in   integer
                                           , product_limit_in integer)
  RETURNS SETOF integer AS
$func$
BEGIN
   RETURN QUERY
   WITH featured AS (SELECT supplier_id FROM products LIMIT 2)
   SELECT supplier_id
   FROM   featured

   UNION ALL
   (
   SELECT p.supplier_id
   FROM   products p
   LEFT   JOIN featured f USING (supplier_id)
   WHERE  f.supplier_id IS NULL
   LIMIT  product_limit_in
   )  -- parens required - or not?
END
$func$  LANGUAGE plpgsql VOLATILE;

• LIMIT can only be applied once in a UNION (ALL) query, unless you enclose the leg of the query in parentheses. You may or may not want to add parentheses.

  • The way I have it, a maximum of product_limit_in rows are returned in addition to the "featured" rows from the CTE.
  • If you remove the parentheses you get a maximum of product_limit_in rows total - meaning that even "featured" products may be discarded.
    Related: Optimize a query on two big tables

• Either way, don't ORDER BY the outer (combined) result before you LIMIT, if you can avoid it. Postgres can optimize the query very efficiently and just stop evaluating once enough rows have been returned (possibly fetching tuples from the top of a matching index). That would not be possible any more, which can make a huge difference in performance.

• Using LEFT JOIN / NOT NULL to exclude featured rows from the second SELECT, which is probably faster than NOT IN and does not carry "surprises" when dealing with NULL values or empty results.

  • Select rows which are not present in other table

• In Postgres (as opposed to some other RDBMS), you can refer to p.supplier_id and f.supplier_id after joining with USING (supplier_id).

And yes, the CTE is only evaluated once:

A useful property of WITH queries is that they are evaluated only once per execution of the parent query, even if they are referred to more than once by the parent query or sibling WITH queries.

Bold emphasis mine.