Question: Default percentile of 95% in HausdorffDistanceImageToImageMetric

I wonder if someone could highlight to me the reasoning behind the default percentile of 95% in the Hausdorff distance (HD) ImageToImageMetric.

Intuitively, I would expect the HD to reveal me the largest error e.g. to denote outliers in a image segmnentation task. But with a default percentile value of 95%, the results show something else (I am not quite sure what). Is there any reason behind disregarding the 5% most distant errors?

I personally think a default percentile value of 100% would lead to less surprises and is more consistent with the classical HD definitions (e.g. http://en.wikipedia.org/wiki/Hausdorff_distance).