Just guessing here, but this sounds a lot like what we found trying to select data from a Netezza data appliance (Netezza is a competitor of the better-known Teradata). It was taking f o r e v e r to get 180,000 rows back from a 3.5 million row table. Since Netezza is normally very fast for queries even with multi-table joins, it looked as though the trouble was in the way the MS SQL box and Netezza talked and matched data for the join, apparently sending data back and forth across our WAN. We ended up punting by establishing a new key table on Netezza, sending the 180,000 relatively small rows over en masse and doing the join entirely on Netezza. There may have been a more elegant solution, so if you've found one for your Oracle connection, I'd love to hear it.