Node monitors flag that node down for all pools, regardless of how those individual pool monitors are configured. It's also worth noting that the node monitor will disable all pool usage of that server without all of the individual monitors having to observe a problem, which is useful if your pool checks are set to high intervals.
Pool monitors just flag an individual member of the pool down, but have no impact on other pools.
Most of the time you want to stick with monitoring pool members, yes. The extra flexibility to completely remove a server is there if you need it though.
A use case would be something that throws the entire reliability of the server into question. A HTTP test that indicates that all services are tainted, a SNMP monitor (I seem to recall that these don't have to be pointed at the node itself) that indicates an upstream problem, and so on.