Chris, what version of Oracle are you runing?

I'm sure wWhat I have written about occupying space in buffer cache is true for Oracle7, I'm almost certain taht it is more or less the same for Oracle 8.0, but I know they've made significal changes regarding the LRU algorithm in 8i, so my remarks might actualy not be valid for 8i. I'll have to check that myself (when the time permits...). Quite some time ago (it must have been 7.3.4) I've done some tests. I've totaly filled my buffer cache by scaning large portions of various indexes and tables via index-read ROWIDs. Then I've performed a full scan on a large table that have had no block in buffer cache. After this I checked the contents of buffer cache (V$BH) and found out that only a smal portion of cache was filled by blocks of a table I've scaned. As far as I remember I haven't performed full scan on freshly started database, so it might be true that if large amount of buffer cache is free the table scan will simply use all those free buffers.

In any case I don't think on busy system Oracle allows that one huge table full scan could simply flush out complete buffer cache (if the has not been declaerd as CASHEd table). It simply wouldn't be logical performance-wise. They have preventing this to hapen in earlyer release, so I don't think they would allow it in 8i.

But as allways, I could be wrong....