On dual GPS with GPS_AUTO_SWITCH == 1 it gets very confused with the compass and does strange stuff, for me it started oscillating on the yaw axis, whenever switching GPS. It also almost never uses reach because it compares the hdop from the standard GPS with the vdop of reach (bug explained above), and ends up almost only using the standard GPS because vdop its typically bigger than hdop.
With GPS_AUTO_SWITCH == 2 (GPS Blending) then it uses Reach but blends in the other GPS because Reach stops delivering accuracy information as soon as it gets a RTK fix. So when reach is fully locked the autopilot actually thinks that the solution quality degrades and starts giving more weight to the other GPS, again a reach bug.
This method did conceal/diminish the yaw oscillation caused be the ground course calculation bug.
Missing speed accuracy information also makes it harder for the autopilot to blend the two GPSs. It would be nice to see it implemented. The ERB driver in the Ardupilot code already flags it as valid although it is not. But that will stop being a bug as soon as Emlid implements it on the reach side 
When using reach as primary and single GPS source (against Emlid recommendation) should work, but the EKF will get confused by the buggy hdop/vdop and missing accuracy information, and give too much weight to the IMU and too less to the GPS corrections. I must admit I did not try this combination. I only tested the two other ones above.
Try it out, and tell us how it goes