Better connected/faster servers reduce it a lot.  3 orphans in 24 hours is seems extremely high given the size of the pool.  Orphan rates should be <1%.  Obviously a given sample may have much more than 1% just due to lack of # of rounds being evaluated.  Maybe organofcorti can shed some light on the probabilities.

EDIT:  2 of the 3 orphans were never even seen by Blockchain.info...that points to serious server issues.  These blocks were either never broadcast, or were broadcast well after the prior block (meaning no connected peers accepted it and rebroadcasted).  The timestamp on 265441 was more than 1 minute after the competing block was seen by Blockchain.info.  The timestamp on 265419 was almost 4 minutes after.  This may not be the real delay due to ntime rolling/slush's poold configuration/where the table get it's timestamp from.  But those are both blocks that were not even seen by the rest of the network apparently.