Additionally, I don't think the current setup is ideal for testing:
We should use a local source build as it is much more common, unless testing remote case specifically
We shouldn't use relatively big go build for this, unless we want to have a test case specific to go (eg. with cache mounts). Pulling down and extracting the Go image and compiling the source code skews the test results for such builds. If the task is to test high parallelization of buildkit, most of the resources should be used by buildkit itself and not external tasks.
Make sure tested codepath is longer and doesn't end without trying to export the build result.
Local run: