There should not necessarily be a difference because you use the same estimator if you want to improve your results you should use another statistical program, otherwise you should perform the mahalanobis distance to evaluate the atypical data and eliminate them, which would be those .000 values and then evaluate the model with a better distribution of the data and better SEM fit values.

Thanks for your response @Cristian Ramos-Vera!

I have already eliminated any outliers via mahalanobis distance before I run the EFA and CFA. My valid sample is N=250, and I am trying to estimate a model with 9 latent factors and 46 items.

I am still confused how the results can be the same when one estimation methods assumes normality (ML) and the other does not (bollen-stine)? I just feel this should be reflected in the model fit indices? Apologies if this is a stupid question.

Use the ULS estimator.