Yes, that's it, and idxA is not unique. Pretty much all the text/blob data is stored (and de-duplicated) in specific tables, all the other tables are thus only holding relatively few integer IDs, numeric values or timestamps. I recently added two composite indexes on other tables to speed up some critical queries (20 GB for each index, which is why they had not been added before...), with those indexes table B becomes unnecessary in many queries, but this placed table A straight back into many critical queries... After running the analyzer, I guess a couple tables and indexes are headed for the denormalization chopping block, I had not realized the indexes had grown so large. Thanks