There's always and inevitably a lot of misconception or even urban legends around anything technical in wide use of non-technical "audience".

The most common USB-related one is the percieved equivalence between "USB 2.0" and "high-speed".

The non-shielded cables (or their kin, the cable having a single, and usually extraordinarily thin, wire, instead of proper shielding) are around due to the fact, that most low-speed USB devices and many full-speed devices in typical "domestic and office" use are most of the time capable of working with such cable. The point is, that the ground and VCC wires in close vicinity of at least mildly twisted data pair works as a fairly good shield.

But here the point is slightly different: the problem is most probably not with noise induced *into* the cable; rather, that excessive spikes induced through the optos' (and the DC/DC if you haven't already gotten rid of it) parasitics, having no better path, flow through the "working" GND wire, thus inducing spikes also in data and VCC. The shield is here aimed to provide a shorter path to true ground, avoiding sensitive circuitry.

Of course this is a double-edged sword, as the shield can in other circumstances act as a capable antenna picking up EMI itself. Also, as I already might have mentioned, some PCs don't connect the USB shield to chassis, and/or don't ground (earth) the chassis. There is no perfect solution, only compromises.

JW

PS. The transils from data lines to GND I mentioned, did you try them? Properly polarised Zeners will do partially, for experimenting purposes, too.