I don't go over the things too often, but I've never noticed this with a wired unit (haven't had my wireless long enough to really notice this, since I don't do that much city riding).
The interference could work two ways. Could be interfering with the wireless radio (which I suspect is ANT, a sort of poor-man's Bluetooth), or it could be that the pulses from the induction loop are triggering the reed switch pickups (or perhaps they use Hall-effect sensors in the newer units).
Whatever the interaction is, it is a bit weird -- Wikipedia says that the typical loop frequency is 10 kHz to 200 kHz. But unless it's pulsed it's hard to see how it would affect the sensors themselves, and ANT operates at 2.4 GHz and should be pretty much immune to the loop frequency.
But the way you say it behaves -- getting "stuck" at the last valid speed reading -- is even more bizarre. My understanding is that the ANT transmitter pulses for each turn of the wheel, and one would expect interference to register as regular pulses, generating a speed that's fixed for a given loop but not related to your former speed.
This points to the 3rd possibility that the cyclometer employs a phased-locked loop of sorts to filter "noise" from the speed pickup, and the induction loop is simply overloading the cyclometer receiver to the point that it detects continuous noise. In this situation the phase-locked loop would tend to maintain it's former reading, and you would likely see the phenomenon of a "stuck" speed reading.
In any event, more and more of the intersections around here use cameras instead of induction loops, so such interference is getting less likely (for me, if not for you).