Your graph is showing gradual convergence / divergence between the sensors; this points towards a thermal effect not electrical. Out of the top of my head, I see three possibilities.
If the last photo is representative of your setup, I would say that although it's out of the case, you still have the BME280 positioned very close to the LEDs, and also in the path of the airflow in/out of the case. To rule out heating caused by the LEDs, I would position it at least a good 20cm away and even put a physical barrier between it and the sculpture.
If your software samples at a different rate when the LED is on/off, this could change the amount of self-heating in the sensor.
Your graph has the LED on/off periods at fixed times of the day. That makes it correlated with human activity. So it might be that there is more air turbulence when humans are moving around and that dissipates more of the BME280's self-heating. The SONOFF sensor would be less sensitive to this if it's protected in a case.