I was a bit incomplete in my post. 225.7(B) applies to outside branch circuits for lighting, and 215.4 applies to feeders. But article 210 is the relevant article, and it doesn't mention common neutrals at all.
The idea of a common neutral just 'feels wrong', but thinking about it, if properly sized it would offer some interesting benefits.
1) Fewer current carrying conductors in a pipe for de-rating calculations.
2) Lower voltage drop, since as wires get thicker, the ampacity increases more slowly than the cross sectional area. A #2 wire has 4x the cross section (and thus 1/4 the resistance) of a #8 wire, but an allowed ampacity of 2.3x that of a #8 wire.
3) More robust shared neutral connections. This is just a guess, but I don't see you series wiring with a number #4 into a 20A receptacle. You would have to pigtail using some sort of wire connector or tap for large wires, and this would mean that you are unlikely to break the neutral circuit...breaking the neutral with one half of a multiwire live is IMHO the biggest danger of multiwire circuits.
The downsides are probably cost and confusion.
1) Confustion: While a competent electrician should know how to deal with a multiwire branch circuit, a common neutral circuit is a strange beastie, and I wouldn't expect most electricians to have seen one. As mentioned, the authors of the MCGraw Hill book believe them to be a code violation, even though I couldn't find the explicit violation.
2) Cost: You end up using quite a bit more copper for this installation. The neutral conductor has to be sized for the maximum possible unbalanced load...and as mentioned above, the cross section of the wire increases faster than the ampacity. In the example that Bob gave, which would be cheaper: 4 #12 wires (the 8 20A circuits done as normal 'multiwire' circuits) or a single #4? In the three phase example, it is 3 #12 conductors versus 1 #4...note that a #6 wire has the same cross section as 4#12 together.
-Jon
