Steve,
I did not increase the separation to 4kHz as the audio filters in HDSDR makes that impossible
As the HDSDR uses an i/q audio (192kHz bandwidth) to steer the output of the SoftRock when you offset TX away from the LO you can get intermodulation products at the other side of the LO
The LO of the softRock was 7.1MHz, TX was at 7.104MHz (USB) with the fundamentals at 7.104300MHz (300Hz) and 7.106500MHz (2500Mz) , The LO of the I/Q mixer was at 7.1MHz so what you see to the left of 7.1MHz is an intermodulation product originating in the I/Q mixer causing the mirror suppression to fail so it appears at the other side of the LO
The same (equally strong) products appear to the right of the LO around the 300Hz and 2500Hz signals but as the original signals are not visible at the other side of the LO these do not mask the intermodulation products at the other side of the LO.
