Regression estimator versus sample mean — an asymptotic, not universal, comparison

Let the variances of the regression estimator and the sample mean, both computed from SRSWOR of the same sample size, be denoted by $V_{Reg}^2$ and $V_{SRS}^2$ respectively. Then: