I'm guessing; but it looks like celery worker_concurrency defaults to the #CPU cores on the K8s node. Fiddling with

config_source.worker_concurrency

as below I was able to get a single Celery worker to run multiple solids concurrently on a 1 VCPU GKE node. (Not sure this is actually makes things faster; but it does seem to "work")

runLauncher:
  type: CeleryK8sRunLauncher
  config:
    celeryK8sRunLauncher:
      configSource:
        worker_concurrency: 4