My current preference is to use a third party help authorizing system to generate HTML based help.
We then use a WebBrowser
to display this help as needed. The authoring system we use makes it fairly easy to extract out a single page from the main help (each "topic" is a single HTML file, and can be included with full contents or not as desired).
Granted, this definitely felt like a bit of a nasty hack at first - but once we wrote the basic plumbing (some attached properties for xaml to specify attributes for context location and add behavior to trigger help, etc), it's fairly clean.
One very nice advantage to this approach, however, is a single help system build works perfectly in all contexts - we can include the documentation online, expose it locally for use in a browser, and use it with context from within our application directly.