I have no clear explanation yet. Crosstalk would be just another word for the observation but doesn't clarify the signal path. A ground loop actually may exist. My first question is, if the channel sourcing the crosstalk is actually overloaded, so a distorted signal is present in this channel? The headphone amplifier would be first candidate for overload, cause it has the highest gain. By reducing the possibly overloaded headphone amplifiers gain for trial, you could check the assumption.
If the amp verifies as crosstalk source, you still have to find the path. By reducing gain of individual stages starting at the frontend, you can try to identify the receiving node (could be more than one in case of a ground loop). Measures, that may be also tried is adding bypass capacitors to supply and virtual ground nodes, e. g. in parallel to R5, R95, R105. By the way, in the schematic, there is a floating virtual ground node at U2.5, U2.10, U3.5 and U3.12. It would cause problems if actually existing in the circuit.