FWIW, Craig B posted a link to a very recent article from Anders that seemed to suggest that in-memory datasets will soon be a feasible approach, which does tend to imply some sort of disk-memory spanning. I think perhaps you're doing quite a bit of implying. ;)
By the way, do you mean Craig Boyd or Craig Berntson? What link/quote do you mean?
Just because someone talks about "in-memory" datasets (which, in itself, we have today), doesn't necessarily mean disk-memory spanning. If you want to lobby for that, fine, but I would think you'd be looking for specific and explicit statements to support your it.