Pitting VESC Six EDU against a Castle Creations ESC - the Castle is winning?

I am evaluating VESC using a VESC Six EDU. The application is propulsion for a solar powered fixed-wing UAV, so efficiency is the primary metric on which I am evaluating VESC.

The motor is a Scorpion S-5028-220kv. When VESC performs parameter identification on the motor, it reports: R=12.5mOhm, L=18.59uH, Ld-Lq=5.62uH, lambda=3.651 mWb. On my scope it has very sinusoidal BEMF when spinning with a drill.

The motor is mounted on a thrust stand with a propeller, and I am comparing to a Castle Creations Phoenix Edge 120HV using the following methodology:

• With the VESC connected, run the motor up to a fixed RPM (17250 eRPM). Measure the resulting thrust (~600 grams) and power input (~90-100W)
• With the CC ESC connected, run the motor up until the thrust stand measures the same thrust (~600 grams) and measure power input (~80W)

This is a really surprising result to me, as it seems that FOC would be a more efficient commutation strategy, especially considering that this motor produces sinusoidal BEMF. I've been tweaking every setting I can find, trying to improve efficiency.

Things that have helped somewhat:

• Increasing Zero Vector Frequency to 60kHz
• Enabling MTPA (IQ Measured)

Things that have not helped or have harmed efficiency:

• Tuning or detuning the current control loop
• Tuning or detuning observer gain
• Increasing or decreasing flux linkage
• Changing Current Controller Decoupling from disabled
• Changing Observer Type from mxlemming_lambda_comp
• Changing Observer Offset from -1 to anything from -5 to 5.

Everything else I have tried has failed to improve efficiency. The Castle is still winning by a long shot. This makes no sense to me.

Things I can still try:

• Attempt to measure losses in the ESC to determine whether the inefficiency is due to poor commutation strategy (software) or conduction or switching losses in the ESC (hardware)
• Attempt to measure the absolute efficiency of the castle ESC and scorpion motor combination from the DC power input to the motor shaft. If this thing is already 95% efficient, I am wasting my time trying to make it better!
• Get an encoder and measure rotor angle error and/or try sensored FOC