The following article compared the two and concluded that Shapiro Wilk is better than Shapiro Francia in most - but not all - cases tested (in agreement with other findings reported in the literature). I would personally go with Shapiro Wilk.
Article Comparing the performance of normality tests with ROC analys...