The Capon or minimum variance distorsionless response (MVDR) beamformer is a prominent example of spatial ltering structure in sensor array signal processing. Typical implementations of this beamformer are based on a regularized or diagonally-loaded version of the sample covariance matrix estimator. Conventionally, the performance evaluation of the beamformer relies on a measure of the signal-to-interference-plus-noise ratio (SINR) at the lter output. In this paper, we establish a central limit theorem characterizing the output SINR performance of MVDR beamforming implementations employing diagonal loading for both spatially and temporally correlated Gaussian observations.