Here's my guess..
We should keep in mind two different strategies for panning:
level difference between L and R
delay difference between L and R
When you say "panned wide", I suspect you're hearing the effect of level panning.
There's a similar way to think of mic techniques.
Coincident mic techniques (like blumlein, X-Y) have the advantage of being essentially phase-aligned, but that means the stereo image won't be attributed to delay differences. Also, the polarity patterns of these mic techniques are usually intended to overlap less. So this way, when an instrument is picked up strongly by one mic, the other mic is probably not picking up as much of that signal - i.e. level differences.
Spaced techniques (moderately spaced, tree. maybe not far-spaced outriggers though) may pick up most instruments across the stage with roughly the same level. So in this case, it's the delay info between L and R that helps to position the instrument.
So... from the "Think MIR" document, as far as I can tell, one ambisonics array (4 mics) was used to represent a mic position. But the awesome part is that ambisonics allows the mic to have arbitrary rotation and arbitrary polarity pattern. So to have a stereo signal, a single ambisonics array can be used twice, each time with different parameters, to represent L + R mics - but please note, we'd really need Dietz to confirm this. I could be very wrong.
But this is where the big caveat is - the way ambisonics is used to achieve stereo mic techniques, this means only coincident mic techniques can be achieved. So how can we get a spaced mic effect with MIR? (a) MIR does have an advanced option to offset the distance of a mic, so that it's possible to emulate a spaced mic pair. But for me personally, I've never felt that this distance offset had the desired effect. To do it really correctly, individual reflections within the same IR would have to be warped with different delays in order to accurately simulate a different mic position, and that would probably require a costly complete acoustic simulation. (b) A more practical way would be to have captured multiple mic positions that would represent where L and R spaced mics could be positioned. But all venues I looked at in MIR did not capture specific L + R mic positions like that.
Which leads to (c): I imagine that the advanced offset option probably does accurately change the delays in the dry signal. In that case, it might make sense that "clustering all the instruments together on the stage" - which you saw in a composer's template - probably could be making decent use of dry signal delays to provide more of a sense of space?