More and more I find it important to differentiate as much as possible the direct-front-arriving-sound from the diffuse sound which arrives from all directions essentially equally. I increasingly see this as one of the more fundamental aspects which OMT exploits. Even though the diffuse sound arrives from all directions equally, if we are to try to pick it up to the exclusion of the front-arriving direct-sound, we need to point the microphones further away from the front in order to skew the balance as far as possible toward the diffuse stuff. Not because the diffuse sound is any stronger from that direction, but because the front-arriving direct sound is sufficiently less strong there. And that strong front-arriving direct-sound is not just coming from directly forward at 0-degrees, it is actually arriving from a pretty wide angle across the entire front quadrant.
The fundamental problem I see with upward facing microphones is that regardless of the microphone pattern used, they will simply pick up too much front direct-sound. The sideways-facing supercards I use as my primary Left/Right channels have a loads of front-direct-sound in them.. much more than most folks realize. They need to, because with surround playback they are directly feeding the Left and Right front speakers. Similarly, just like my sideways facing L/R pair, an upward facing supercard would also be 90-degrees off-axis and pickup about the same amount of frontal-direct-sound.
Sure, direct-sound arriving from straight ahead is about 9dB down in a 90-degree off-axis supercard (which isn't really that much), yet the direct-sound from in front isn't arriving from just dead-ahead. It is actually spread out across the front quadrant. So what we should really consider is something more like the the average sensitivity across the front quadrant for any microphone facing 90-degrees off-axis. Sound arriving 45 degrees off-axis to a supercard is only 2 dB down. Even with an figure 8 the sound arriving from 45 degrees off-axis is only -3dB down. A figure-8 has a nice deep null at 90-degrees off-axis, but it's simply not wide enough to be effective at this.
If you ever use a center Mid-Side pair in combination with a rear-facing cardioid or supercard channel, you can hear this for yourself by soloing just the Side channel, then the rear-facing cardioid channel, and comparing the two. Listen to how much front-direct-sound is in the Side channel in comparison to the rear facing channel.
Because of that, I think any microphone intended to exclude the front direct sound as much as possible needs to be a supercardioid pointing at least 130 degrees away from the front, or a cardioid pointing even further away from the front. And I feel this is just as important when the intent is mixing to 2-channel stereo as it is for surround.
Although I see getting sufficient direct/diffuse differentiation as most important, there are other less important things the rear-facing microphones can provide.
Picking up the direct-sound components of audience reaction and room reflections arriving from behind is very nice for both 2-channel and surround (more so for surround where it anchors things with discrete sounds from directions other than the front), but is a far second to getting good diffuse sound with minimal frontal-direct sound in it. And many times there may not be much if any good direct audience reaction stuff or good room reflections back there anyway, depending on the venue and recording position.
So somewhat upward-facing when backwards-facing isn't practical. Otherwise I see it as an unnecessary extravagance, even if one was actually recording for surround playback that includes overhead speaker channels, where I suspect it would be more fruitful to use ambience-extraction/decorrelation techniques to generate those extra channels from the existing surround channels.