The current design of the RANDR CRTC implicitly assumes that all CRTCs share a single allocation of framebuffer memory.  This is certainly the simplest thing that could possibly work, and it's very efficient in terms of rendering overhead.  However many configurations are approaching a point where the total desired pixel area exceeds the allocation limits of a single CRTC.

An example is instructive.  Intel 945 hardware has a CRTC pitch limit of 2048 pixels.  However, it has two CRTCs, and they can each be pointed at different allocations.  It is not possible to satisfy two adjacent non-overlapping 1920x1200 monitors within a single 2048x2048 allocation, but two distinct allocations would work fine.

Implementing this requires that rendering be broken up across allocation boundaries, appropriately translated within the screen's coordinate space.  Logically you can think of this as having multiple pixmaps backing the root window.

With the introduction of the Composite extension, pixmaps can already have "screen coordinates", so much of the translation work is already in place.  It just remains to implement the various GC and Render wrappers to do the iteration.