To do an accurate test- did you replace the driven speaker (L) with a dummy load so as to eliminate any possibility of the mic picking up sound from the driven speaker?
Yes IIRC the test was repeated with the driven speaker both in and out of circuit. Geometry and windowing were also on our side.
On site RMS voltage measurements with a quality DMM on the speaker conductors themselves corroborate this @ ~ -30dB crosstalk, and was replicated at the shop with a similar length bundle of multiple THHN pairs, resistive load banks and an HP analyzer. Twisting the pairs at the shop bought a crosstalk reduction upwards of 10dB . I suspect you could repeat this experiment and see similar results.
Judging by the HPF-iness of the coherence mask, I suspect the mechanism in this case is capacitive in nature.