Rather than trying to run a bunch of tests. Does anybody know what is less resource intensive?
Bindings, KVO, or Notifications? Has anyone tested to see as well?
Chuck's answer is quite right as far as notifications vs. KVO goes, but I'd like to expand on the "Bindings are KVO + KVC + ..." part he mentioned because that's where the real pain in the ass can be. Key Value Coding seems like it's the more important consideration here, since you can't use Bindings without it. If you're worrying about performance for good reasons, you'd do well to note the cost heavy use of KVC can incur.
I would say that any circumstances that call for heavy querying as a result of a single action (example: asking a few thousand objects for the results of multiple key paths each) would be an indicator that you might want to avoid KVC (and Bindings by extension). Especially with long key paths ( -valueForKeyPath: vs. -valueForKey: ).
I've run hard up against this myself and am now working to eliminate this part of my architecture as a result. The relatively minute cost of Key Value Coding can seriously add up when you're asking 16,000 objects for the result of half a dozen long key paths as a result of a button press (even with NSOperation/Queue). The difference between using KVC and good-old-fashioned [[object message] message...] calls can mean the difference between a few seconds and well over a minute or two on large jobs. For me, the very same query calling the accessors directly (like [parameter name] or [[parameter variable] name]) represented roughly a 500% increase in speed. Granted, mine is quite a complex data model with a large volume of data for a typical document.
On the other hand, if many of your app's single actions affect/query one or a handful of objects and are mostly Key Value Observing-oriented (ie, change a last name and have it update in a few views at once), its simplicity can be akin to magic.
In summary: if your app queries/updates large volumes of data, you might do better to avoid KVC and Bindings for the query/update part because of KVC, not because of KVO.