Except that's not what I'm referring to. As with any cost based optimiser, it's only as good as the statistics it has to work with, and if you've finally carefully tuned your database with an appropriate histogram on a critical table, and found that after an upgrade they "tweaked" the CBO to fix a statistics calculation then you are right back to looking at explain plans to fix your performance issue.

Sometimes you just aren't going to pick up this sort of thing in testing as you'll only know about it under full production load. Their CBO is pretty powerful, but by and large it's often difficult to know why it makes certain decisions. And that's even when you use their full suite of profiling tools!

Having a poorly thought out query planner as the cornerstone of their product is a sign that the product is not that good.