When the software side and RAM prices work together to actually enable the use of this kind of raw power, look out. You would be totally cooking with gas. With this kind of memory, one could still use a streaming sampler architecture to enable huge loads, but with sufficiently large preload buffers so that you would be hitting the disk fairly rarely.
If we were talking about VSL, for instance, it could manifest itself in much more complete performance logic--loading much bigger instruments with more smarts and more articulations into fewer MIDI slots, so that the performance simplicity would go down without losing all of that wonderful recorded material in the tradeoff.
I think the big payoff is going to be that a lot of stuff on the market today was recorded very broadly, much more broadly than can be conveniently mapped and accessed due to hardware limitations. But the recording practices are extremely good. So (and thinking mainly of VSL but also of several others), when a certain threshold of hardware performance becomes mainstream, these tools will have the capability of being reinvented--much like Dave has reinvented the Miroslav libraries by going back to the original material and remapping and reconceiving for today's highest technologies.