I think play from RAM has to be the way to go. In fact anything that can speed up generating real track parts significantly is just such a workflow enhancing thing - that efforts should go into that considerablyl. time is Monet and waiting around kills inspiration. period.
And because PG would control the API and protocol between the plugin and server ( BBM4 ) it means PG can do anything they like - and are not limited by following a 3rd party API ie ReWire.
This is what VSL done with for its symphonic samplers. A shell plugin that talks to service that does all the magic (and is a VST host by itself).
But not so easy to pull off. Yet another area for bugs and roadblocks. If I wanted to connect BiaB with a DAW I'd go for ReWire and live with the fact that it is a loose coupling and not a seamless integration.
I hope the PG team realise that the plugin isn't just a "nice to have" - something to placate a certain subset of users - but could unlock a huge new market- and more than that - make BIAB become a solid staple of modern music production in the same way Ableton and Native Instruments and Protool have been since the late 90s.
I agree that a seamless integration of BiaB as a plugin would be a killer app. If the plugin were done right no one needed yet another mini-DAW. I'm speaking of RealBand.
Till it becomes a staple some work has to be put into the workflow for the users, the user interface design and the graphics design.
I had hoped that the plugin will be a one-stop-shop for fast music production. Seems that I have still to wait and hope.
Masi