I've been doing a fair amount of work with Oracle Text in 9i and especially with indexing PDF's. I'd been struggling with performance of searching as the index was added to.
A context index is structred as below (This is taken from the Oracle text manual)
"Word" DOC1 DOC3 DOC5
where the word contains a list of documents containing that word. However when you add a new document to the index you get the following
"Word" DOC1 DOC3 DOC5
"Word" DOC9
"Word" DOC10
"Word" DOC11
As you can see a new row is added each time the index is updated, and this will continue as the index is sync'd with new documents. This of course slows searching of the index due to the fragmentation.
I have though managed to stumble across the ctx_ddl.optimizeindex method which tidies up new entries to the index by removing duplicates so taking the above example becomes
"Word" DOC1 DOC3 DOC5 DOC9 DOC10 DOC11
Means a more lean mean indexing machine :-)