DP version? If so, perhaps glitches in the hall signals, induced noise from phase wires since they run parallel and unshielded in most setups. Shielding the halls might fix it, or perhaps adding very small (0.1uF or less) ceramic caps to the hall lines at the controller PCB. (acutally only needed on the hall line that's running the CA speedo line)
SA version? Have sometimes had problems when magnet was in same general radius as cover bolts or disc bolts; those would sometimes false trigger the reed at higher speeds, presumably because they had just enough magnetism to retrigger it after it had been triggered and was just disconnecting after the magnet passed, whne the boltheads passed it and retriggered it. Would be intermittent and inconsistent, most of the time this happened to me. Moving the sensor a bit usually fixed it, sometimes requiring moving the magnet and sensor to a different radius than any of the bolts.