NOTE: The plot below is somewhere between misleading and wrong. See reply.

Attached is a plot showing the following (all of these were taken with the laser shutter closed in order to focus on electronics noise):

• Voltage noise out of the mixer with the PD connected to it
• Voltage noise out of the mixer with a 50-ohm terminator in place of the PD
• Noise directly from the PD. Note: the raw data for this was divided by two to account for the change in transimpedance gain when going from Hi-Z (Agilent) to 50-ohm (mixer)
• Measurement noise floor

The "mixer output" here is actually the output of the lowpass filter after it.

The noise from the demodulation seems negligible in comparison to the dark PD noise, with the exception of a broad peak at around 49 kHz that appears to come from the mixer.