Thank you David, hopefully allowing high jerk for travel moves will resolve the issue.

I'll try and source a 3mm bowden tube. In the mean time I'll turn off the pressure advance feature for now. I just printed a floppy cleaning kit someone asked for but the filament lost traction for about 10 layers before picking it up again due to the extreme back and forth motion of the extruder motor, especially with narrow infills like 2-5mm wide.

Especially the turn tool got out with under extruded layers, it was easy to snap them in two parts. I've reprinted it without PA and came out fine.

No change or improvement in the pressure advance department unfortunately. The same k value is indicated.

I've tried a bigger print. The infill makes the extruder go back and forth with quite a bit of noise, clacking back and forth.

I'm not sure this is the way to go for me at least. Does the noise improve at lower k values? It wears the filament down too, with the PA calibration the extruder lost grip on the filament. Luckily it was near the end of the print but it wouldn't be good for other long prints to fail half way.

I may try an older firmware to see if the travel/extrusion move jerk is back on how it performs.

I've adjusted jerk to 1000mm/s from 600mm/s and started skipping steps. I upped the motor current to 1800mA (was 800mA) and stopped skipping steps. It makes a heck of a racket when doing infill. And the gap is still there unfortunately. But the speed is about the same now as without pressure advance.

I'll try doing the pressure advance calibration again since the backlash may have changed since tightening the gears, there's no play in them now.

I've increased acceleration, velocity and jerk (slightly) but it doesn't seem to improve much. The left most seems to be the best. At the top you can see the outer perimeter near the hole isn't as thin as the other two. I'll try bumping up jerk as you suggest, its at 600mm/min, I'll try 2400mm/min.

The bowden tube diameter is about 3.1 to 3.2mm when I scan a cross section and measure it in photoshop. With the calipers it's between 3 and 3.1mm. The ABS filament is about 2.85 to 2.90mm (brand: RRW Dutch ABS).

The push fit connectors don't show any motion. I use 3d printed clips to secure it in place. There's a bit of wiggle room between the extruder gears though. I've attempted to tighten them but ended up with skipping steps. That was before the duet upgrade so I could try it again. The drivers provide much more power to the stepper motors.

Bowden connector on the x carriage:

Bowden connector on the extruder side, there's no play in the tube nor any movement when printing/retracting un retracting:

Here's the same test with setting k= 2 to 6. The third layer looked good to me, which corresponds to k=2.24. I have this set in the config.g file now.

I've reprinted the 20x20x2 slab. To the left is the 2.24 newly printed, and compared with the 0.2 one printed earlier. A clear improvement and the blob on the top is almost gone. Instead there's now a hole, probably due to the pressure not being able to come back up quickly enough. The infill cornering is improved as well, I can now set the infill/perimeter overlap higher to fix this. There's a print speed penalty though as the acceleration is now adjusted to the speed/acceleration the extruder can handle. I will tune that further first and see how it will perform. I did notice that the infill looked different, in the center it seems wider and flatter and near the perimeter it's narrower and rounder. Coming into the corner looks different than coming out of the corner. The left one looks more consistent apart from the blobs at the end of each infill cornering.

I've looked in my failed prints bin and found a couple of marlin 1.1.9 prints, but I'm not sure what settings and what stage of calibration it was. I had to calibrate motor current at some point, not sure if it was before or after motor calibration. My guess is the left one is pre motor calibration and the right one post motor calibration looking at the extrusion line spacing, left is irregular and right is more regular. I don't know if jerk was changed either. At least I've gotten rid of the corner blob that is still present in the print below.

Here's the pressure advance calibration test with the settings K=0.4 to 2.0. 2.0 isn't high enough, so next test print will be 2.0 to 6.0. Is my bowden setup that bad? Maybe the teflon tube is a bit too wide?

Thanks for the script! Please note, to use the gcode you should use relative extrusions using M83, otherwise nothing is extruded. I think slic3r uses absolute extrusion M82. If you copy/paste startup gcode this could affect the print.

Thanks for the link, I'll check it out. It's not the exact same gcode but the same settings as with the test print. I didn't do any test prints on marlin 1.1.9, only actual prints of things I needed like an extruder holder for the e3d v6 (which is what the last image is of).

Thanks! Before your last comment I printed four test prints as shown below. Clearly the top and bottom is a big difference but at 1.6 the printer is slowing down as if a low acceleration is set. The extrusion stepper motor is doing some wacky back and forth motion on the infill too, though I must say the infill looked fine.
I'll try reducing retraction distance tomorrow. (please ignore the bent prints, I've pried them off while the bed was still fully heated, it didn't warp during print.)

Marlin 1.1.9 seems better controlled, yet doesn't use pressure advance:

Thanks, I wasn't aware of that option. I'll try this in config.g:
M572 D0 S0.2

The bowden tube is about 600mm.