Oily, with lumps of gristle
The Shannon-Hartley theorem (quoting freely from Wikipedia) "establishes a bound on the maximum amount of error-free digital data (that is, information) that can be transmitted with a specified bandwidth in the presence of ... noise interference, assuming that the signal power is bounded, and that the Gaussian noise process is characterized by a known power or power spectral density."
In the case of DIDO, much of the 'noise' experienced by one user of the channel is not random (unpredictable) gaussian noise, but is actually the result of the combined signals from other users. Since DIDO knows (almost) exactly what these signals are, it is reasonable to expect that it could improve on the information limit specified by the theorem, although it is hard to know to what extent. The Wikipedia article suggests as much: "If the receiver has some information about the random process that generates the noise, one can in principle recover the information in the original signal by considering all possible states of the noise process". In practice, the limits are likely to be determined by the amount of true (unpredictable) noise in the channel, the sensitivity of the receivers, and the fidelity with which the transmitters can reproduce the ideal signals specified by DIDO. Tests with just 10 users are not going to establish the real limitations.
Exploting the relative independence of different signal paths between users and access points is also sensible, but there does not seem any reason to expect it to be massively better than the multi-path technologies already in use.
Combining signals from multiple sources to achieve the desired signal at a point is far from new - anti-noise technology has been doing it with sound waves for many years. However, and this is the oiliest point, to achieve this some of the peaks of the transmitted signals will need to be higher than the peaks in the resultant signals (this is obvious if you think about the reverse operation - if two users transmit their desired signals, some of the peaks at a receiver will interfere constructively). Thus to stay within regulatory power limits, the effective power experienced by a user in the channel must be reduced, thus reducing the channel's information-carrying capacity and hence reducing some (all?) of the gains claimed for the system.
On top of all this, DIDO requires a complete lock-in on all users of a given channel in a given area. Unless it can demonstrate huge performance benefits, such a degree of control is unlikely to be acceptable (although perhaps more likely in certain dedicated application areas, as the DIDO brief seems to suggest).