Stats are on.

Computed, with histograms for indexed columns. Indexes also coalesced - and where significant free-space generated, rebuilt.

The thing I was going to mention earlier (but forgot), most SELECTS are done via views that have IN-lists within the code. This is outside vendor programming which we are stuck with for the present.

Consequently, I was of the assumption that we might alleviate scattered reads by increasing index caching.

Taken a look at most table blocks in the buffer cache that were read in as part of a full table scan. For the table concerned have traced full table scans and the (average) multiblock read was 4, 7, 10 respectively. Only the smallest of tables could actually achieve 16 block reads - presumably are they are available contiguously on disk.

T.