An Empirical Study of Accuracy-Robustness Tradeoff and Training Efficiency in Self-Supervised Learning