The overlay dataset concept is interesting and deployed today as part of the CMDB functionality around the sandbox datasets if my understanding is correct - BUT - this is all done using the API mechanism and hence seems to work as advertised. We also tried (many years back when this was first made available) to see how it worked via the UI and got nowhere. We had a number of support calls about it and the advice was 100% consistent with the caveats you have seen from BMC. Basically it is only workable at an API (i.e. programmatic) level.
I didn't attend the webinar (I'll try and catch a recording), but am very surprised BMC are still highlighting this when the practical availability of it is so limited. We had several use cases where the overlay dataset concept, if working in the UI context, could have been really useful, but had to abandon them since we couldn't support an API driven approach to the requirement.
Simple answer is that BMC would be better to present this capability as an API only interaction, OR do whatever is required to make it work once configured, for user interactions. Advertising that it can be used but with unreliable results, isn't a good look.
Thank you Mark Lev for starting this discussion. I just got done watching the same Webinar and was a bit confused as well.
I really appreciate the information Carey Walker. This definitely clears things up. Also, if anyone is viewing this conversation and is interested in watching this Webinar (Which has a lot of great information), I have provided the link.