You need to refine the question that you are trying to answer...
"Are the missing observations normal?" No, they are missing.
"Are all complete cases normal?" You already have that answer.
You could apply your tests to subsets of the variables, which would give you statistics on different subsets of cases. At some point, you will have to concede the unlikelihood of getting a different answer.
You could use multiple imputation and apply your tests to that. Multiple imputation performs pretty well, and if you are going to use the imputed data for analysis, then "Is the imputation-result data normal?" seems to be the relevant question, assuming that normality is relevant to your analysis.