Inferential optimality and ethics in clinical trials with covariates