The answer is that single pass rendering should not substantially lessen SLI's effectiveness at offering performance gains over a one-card setup.
The view-independent work is duplicated in each card in SLI, while the view-dependent work is not. It appears that it's mainly just shadow-related work that is duplicated. In addition to duplicated work, about 1 ms is taken to send GPU2's image to GPU1. Single-pass rendering is going to lessen somewhat the amount of work that is considered view dependent. The exact amount is unclear, but it seems like most of the savings will be CPU related, as the following bullet quotes from the Multiview Rendering (Single Pass Rendering) slide indicate:
"Saves CPU rendering cost. Maybe GPU too — depending on impl!"
The time performance savings that a SLI system will have on a work load identical to that of a non SLI system, assuming no single pass rendering being used, will be 30-35%, according to valve's GDC talk:
But, if the SLI system is rendering more pixels, then double the resolution can be rendered at the same speed as the non sli system rendering its standard resolution. Valve is not using single pass rendering, as is proven at around 9:30 in the demonstration. The amount of performance that VR SLI will provide will likely not be double that of a single card, then, as the Valve presentation suggested, for systems using single pass rendering, but the performance gain should still be massive.
Valve's Source 2 version of SLI-VR, though not officially Nvidia's method, appears to be nearly identical, though coded from the ground-up by Valve.
Unity and Unreal Engine 4 appear commited to implementing all of Nvidia's VR works. Unreal's update map has both single pass rendering and multi-gpu rendering on it's to-do list for the next few months. Unity has only single-pass rendering on their timeline but has committed to implementing VR Works, in general. Valve seems intent on not using any of VR Works, seeming rather to desire to implement their own solutions for Source 2. For example, in the conference linked to above, the guy speaks of coding his own version of multi-res shading, apparently having never even heard of Nvidia's version. They have expressed their desire to never do anything like single pass rendering, due to the difficulty of adapting it to Source 2. In the video above, however, they seem to suggest that they will be developing future VR games in Unity, so all of Nvidia's VRworks may be used by them, unless they're going to avoid them out of principle.