RLamb Attached is a LOG and screenshots of guide settings from last night.
Notice that the frame rates of your 0.5 second exposures had dropped to less than 2 FPS. Looking at your guide pulses, it appears to be over correcting.
The ASIAIR is spending more time sending correction pulses to correct itself.
I would suggest first reducing the Max RA and Max Declination durations to less than 400 ms. At a autoguiding frame rate of 2 FPS, I use 150 ms with my RainbowAstro mount, albeit, the RST-135 has a smaller first derivative of the periodic error curve than your mount (contrary to the ZWO boasts, the first derivative is the important number when autoguiding, not the peak-peak error of the curve, which is important only for unguided tracking).
So, you might try starting with something in the region of 100ms or 400 ms first. And then tweak from there. A 2000 ms pulse (the setting that you are using) allows the mount to be moved by 30 arc seconds (!!!) per pulse; only cheap trackers require that amount of correction.
This will prevent any error in correction to throw the errors off that badly (I typically guide at 0.35" to 0.4" total RMS with the very small corrections that I use).
I also use very small loop gain (aggressiveness) with my mount (around 35% and 40% in ASIAIR). You might want to start at the minimum (10%, I think) and only increase the loop gain (by no more than 5% each time, and then waiting 30 seconds to see the response to each change) when the correction pulses cannot keep up with the periodic error.