Parallel scans using block-range granules do direct reads. I have not yet checked whether the same applies for partition granules.

Yes, direct reads do make the traditional formula for calculating the cache hit ratio bogus. However, there are other ways of calculating the hit ratio correctly as explained in the Ixora web tip on Calculating the Cache Hit and Miss Rates.

It would probably be a mistake to tune up your sort_area_size and 'hash_area_size' at the expense of db_block_buffers. Just because your queries are driven by parallel scans does not mean that they do not make use of the cache later in their execution plans. The cache latches will still be used, although less than might otherwise be the case.

I am just trying to understand the following. Using parallel queries in data ware house situation. The degree is 4. Not using MTS.

Is it true that in above situation direct reads will be performed and the buffers fetched will be kept in each users PGA which will outside of SGA. And if so does that mean all db buffer hit ratios is bogus? And are we saying if we are doing direct reads always as in our situation we should be tuning sort_area_size and hash_area_size ? And will db buffer latches will never be used?