A bit late, but here are my attempts at this scene.
Some notes:
(*) I used max8, since the mental ray version that comes with it seems to do a better job at FG, and FG is essential for this image. Further on, mental ray in max8 supports multibounce FG, which is also essential for a correct result.
(*) Since photon-mapped GI in mental ray does not work with skylight, you cannot use it in this scene. Using it will not produce a correct image, as it will be lacking a major component in the lighting. Apparently, this seems to be a common mistake when using mental ray. To make up for the missing lighting, you have to use area lights at the windows, but settings them up in mental ray is complicated and I didn’t do that.
This puts mental ray at a disadvantage for this scene, as you have to use multibounce FG to get correct lighting, which is slow with many bounces. And to get the lighting more or less correct for this scene, you *do* need a lot of bounces, as shown in the images below. Of course, you can resort to some fakes so that you can use photon-mapped GI after all (e.g. adding an additional blue omni light or similar), but I didn’t do that either.
In V-Ray, you can use the light cache to get the secondary lighting correct and fast, and it does account for the skylight. Further on, similar to radiosity methods, the light cache does not place a limit on the number of light bounces. What you get is the correct final lighting distribution in the scene. At present, there is no comparable indirect illumination method in mental ray, as both photon mapped GI and final gather are limited in the number of bounces they account for. Of course, you can get results that are sufficiently close to the correct solution, but this takes additional time.
Here is an image of mental ray rendering with GI, without the sun, to demonstrate better the fact that GI alone does not account for skylight:

full-res image: Chaos: Industry-leading design and visualization software
Here is a series of renderings with different number of bounces for final gather; as you can see, more bounces (more than 1 at any rate) drastically change the result. I have disabled the sunlight to better demonstrate the effect of skylight:

full-res images: Chaos: Industry-leading design and visualization software
Here is a series of similar images done with V-Ray. As you can see, render times are pretty similar, although the result in V-Ray is a little more detailed. In the last image, the light cache accelerates the rendering a lot, plus it gives you the total correct solution:

full-res images: Chaos: Industry-leading design and visualization software
Here I have enabled the sun and re-rendered:

full-res images: Chaos: Industry-leading design and visualization software
The results for both a pretty similar, but the light cache gives V-Ray a distinct advantage with respect to rendering speed.
Best regards,
Vlado