Hi Ed,
Personally I would not trust the 3.3V power supply setup you've got going there. The current consumption of both the sensor and the "A-side" of the TXB0104 will most likely not be constant meaning the voltage drop across that resistive divider will change resulting in an unstable 3.3V supply rail. Now, the cap you fitted obviously will help but I still don't think it's good solution. I'm not saying it will fix your issue but I would at least try with a proper 3.3V supply for the sensor and the A-side of the TXB0104 - at least to rule it out.

Also, make sure you have a decoupling capacitor on the VccB pin of the TXB0104. As you extend the cable length the capacitance of the cable increases, that capacitance will need to be charged and discharged by the driver circuitry in the TXP0104. It'll need proper decoupling.

Are you generating the 32.768Hz clock for the sensor with the CCP module in the PIC?

/Henrik.