PQO and PSO optimisation problem

Can anyone suggest a good Parallel Query or Parallel Server strategy for the following problem:

I have 7 tables with about 2 million rows each. Table structure is:

        date, id1, id2, value

(id1,id2,date) is the unique key

A set of 7 tables represents one week, and the date column is actually the partitioning column when I UNION ALL a large number of these tables. (i.e. the date is the same for all rows in any given table).

I want to product a weekly aggregate

(id1, id2, value)

using code like:

        create table weekly
        parallel (degree N instances M)
        unrecoverable
        storage (freelists 1 freelist groups 6)
        as
        select id1,id2,sum(value) from


        (

                select id1,id2,value from sunday union all
                select id1,id2,value from monday union all
                        ....
                select id1,id2,value from saturday
        )
        group by id1,id2
        ;

The size of the aggregate table will also be about 2M rows.

Now, a simple query such as:

        select count(*) from sunday;
takes about 10 seconds but the best I have done in the query above is 14 minutes. My target is 3-4 minutes.

The problem, as I see it, is that each sum requires one row from each of the separate UNIONed selects, so there has to be a very large amount of traffic between the parallel query slaves to resolve the problem.

The question (finally):

        Has anyone got any suggestion for
        a)  optimal values of N and M above that might help
        b)  hints anywhere in the code that might help
        c)  an alternative strategy


Configuration:
        Oracle 7.3.2
        Parallel Query Option
        Parallel Server option (6 instances)

---
Jonathan Lewis
ora_mail_at_jlcomp.demon.co.uk