A few ideas:

Use adb input to simulate a keypress event to trigger touchboost.
Try throttling GPU clock speed.
Try not throttling CPU clock.
Try running the tests with real touch input and no throttling.

Ideally we'd have some graphs comparing each of the above (and potentially some cases where multiple approaches are being tried).

We'd like to simultaneously minimize noise, while minimizing how far we deviate from running the tests with real touch input and no throttling.