Hey Dave,
I have only a passing understanding of the math involved, but here's where I'm at. While I generally prefer a bit of near-coincident phasing, I think the usefulness of MS is varying the stereo image after the fact depending on how wide an angle you want to reproduce.
Here's an example of what I mean, from the wildly theoretical realm: according to the AES paper by Dooley and Streicher on MS, a 50/50 ratio using a cardiod for Mid, you end up with virtual XY Hypers at 126.9 degrees. If that's right, Williams charts show this to be about +-44 degrees stereo angle. However with M at 70%, they show virtual cards at 81.2 degrees for a +-93 degree SRA. With S at 70%, the included angle of the virtual XY pair broadens to 155.8 degrees, and the patterns become more bi-directional (larger rear lobes).
So the wild part about MS is that as you vary the ratio of M to S (in post production after the fact), for any given M capsule pattern the resulting virtual microphone's XY angle
and capsule pattern changes!
Looking through their billion charts, depending on the M capsule and the M-S ratio, you get virtual XY cards, hypers, subs at various angles, and even a virtual Blumlein pattern when you go 50/50 with a centre Figure 8
So if you're right up close like BayTaynt3d's situation with sound sources coming in from a wide (180 degree) angle, you can represent that in the stereo image with XY cards or by dialing in the appropriate M-S ratio (which depends on what you choose for M). But with MS if its a live event, yer on stage lip, and the musicians move closer together than you expected, you can virtually shrink the stereo recording angle after the fact to represent that new reality. This also protects against the situation I had recently where the opening band's guitar player set up way outside my DIN SRA, resulting in great imaging of the band - but with this otherworldy, ungrounded acoustic guitar floating around it..
There's another guy who did rediculous rafts of tests with real physical Neumann mics whose ultimate conclusion is that generally speaking the frequency response of an MS matrix is better than the actual XY config (I guess due to the on-axis Mid), but that's a whole other topic