Tuning pipeline table function queries

When using a pipeline table function in a query, it appears that Oracle uses the default statistics for that source (nothing surprising about that). Unfortunately, this often leads it to make some bad decisions about the execution plan. I could of course use the well documented hints to explicitly tell the optimizer which join methods to use, table order, etc... However, I'd rather not interfere with the optimizer any more than I have to. Along that line, I've tested things using the undocumented CARDINALITY hint and that gets things back on track nicely.

     Obviously, using an undocumented feature has it's own risks.

What's your opinion?
Any other suggestions?

--
Richard Kuhler