Choosing the right winsorizing method for outliers

So far, I've only been able to find references to 2 methods for winsorizing outliers: 1) substituting the outlier for the next higher or lower existing value in the dataset that is NOT an outlier, and 2) substituting the outlier for either the 5th or the 95th percentile ("90% winsorizing"). I wonder, though -- wouldn't it make more sense instead to set the outliers to the CUTOFF value that was used for identifying them in the first place? Would that not be a more conservative, and therefore reasonable, adjustment? Is that a known or accepted winsorizing method, and if not, why not?