Percentage noise is uniform noise scaled by the percentage of the altitude at each point. It probably should have been percentage of the maximum height so that it would get, say 10% of the height added to each point rather than needing to figure the absolute value.

To get rid of the excess point noise, there are two obvious candidates: Filter>>Morphological>>Erode (minimum value in local area) and Filter>>Morphological>>Median (most likely value in local area). You could also go for blur, but that tends to kill the high-frequency components that you want for that kind of thing and won't fully get rid of the spikes.

There is also a bug in there that adds too much of a result for some operations. I don't recall which ones at the moment. I hope to get back around to looking at Wilbur again some year, but I have no idea when that might be.