Hmmmm. I think it depends on your application. I agree that full table scans would be more costly if part of the information is stored in-line with the table and part of it is stored elsewhere via a LOB locator. But, who designs applications that do full table scans?

We store XML inside CLOBs with the first 4000 bytes in-line with the table and those CLOBs that are over 4000 in a separate tablespace (on a separate disk). We use InterMedia Text to index those CLOBs. So, searching for anything in the XML column results in a hit on the InterMedia index, not the table itself. Then, for most of the rows ids returned from the index data is returned from the table and in a few cases, where the CLOB is over 4000 bytes, a recursive call is made to get the data via the CLOB locator. Much faster.

So, guess it just depends on your application.

-Christopher