Greg Rahn’s post about The Core Performance Fundamentals Of Oracle Data Warehousing – Set Processing vs Row Processing is so good, everybody considering a migration from a standard RDBMS to a VLDB platform (such as Exadata, as in Greg’s example) should be forced to read it.
To paraphrase Greg: the performance improvements of the new system not only allow to run today’s jobs faster, but will allow you to do jobs that were entirely impossible with the old system – if you’re willing to do a little re-engineering, and throw away old assumptions and ‘optimizations’ that make your code slow instead of fast on the new platform.