Basically any video-editing software uses this technique.
When track audio/data is calculated in realtime, the result is automatically cached in memory.
Until the track-data isn’t changed (or the track doesn’t record/monitor), cubase could just stream the audio from memory when its played back again.
Nowadays computers have a big amounts of memory, and with 64bit there is no problem addressing these high amounts of memory.
Then we have practically endless amount of virtual-cpu power.
It’s a cool idea. I think this is what Digital Performer user.
It wouldn’t have to be in RAM necessarily though now we have SSDs. That would get in the way if it’s fully loaded with sample libraries. RAM is the most expensive to add.
Thanks, imho using the filesystem i/o with such “fluid” audio data wouldn’t make no sense when there is enough RAM. Modern Computers have massive amount of RAM which is currently not used. But anyway, hopefully Steinberg will implement something like this in the near future 
In a big project, even 64GB of RAM might get a good chunk eaten out of it with uncompressed 32-bit 44.1+kHz audio, but that’s my estimation. Normal audio tracks are also streamed from the drive, so I don’t see the difference here.