It's no so much the operators you use, but which operators in combination with which optimising methods Oracle can find/use. Your biggest drawback is having little data to play with - access paths can change quite dramatically as the tables get populated and the granularity of indexes change with the data.
I've seen Oracle running full table scans on <50,000 row tables even though unique (pk) indexes were available. I sat there complaining about the optimiser (CBO) only to discover that such a 'small' table returned a lower overall cost with a full scan rather than with a forced index unique by rowid - but with higher disk reads.
It is quite difficult to monitor and check access paths without a realistic data set to play with.