Simple example. You have a parent child relationship in your data that has a ragged child hierarchy. You need to use different logic on different types of parent records and then on different types of child records that are related to that parent, and even based on the number of child records you find the logic to be executed needs to be different. The non set based logic of a cursor makes debugging easier as you can very easily step through the logic, rather than simply looking at the result of a set operation (or more likely a number of set operations)

While it will be possible to do with set based operations and numerous temp tables - in some circumstances cursors provide a better mechanism, not sure on speed. But clearly if a cursor is perfoming in a 'performant' manner from a user perspective and resources are not constrained on the server, then what is the issue. I have also seen set operations which perform poorly, so let's not necessarily blame the construct.

Again I am not sure why Oracle can make their PL/SQL using cursors as fast as set based operations and MS cannot. Cursors are not evil, and not bad, just a different method of processing data.

Just my opinion, and I will use them where I think there is an advantage in doing so.

Cheers