Yes probably reducing the 8 digit accuracy to 4 digits would give similar effect for 90% of the cases. Did you change the scale on import-export?

I referred to 'noise' from various sources like it comes from say making the morph via simulation. If I make say a belly morph in a sweater that way I will find there are insignificant but non-zero deltas in the sleeves.

I am considering to add a sort of filter function where morph deltas of vertices less than distance x from the area of interest are written with weight factor 1.0, and outside that area with a factor decreasing gradually with distance, to be filtered by the lower limit where appropriate. Something like the falloff zones of joints. Default zones would need to be defined only once for a body morph normally.

From there, route would be open to very interesting tools for content development, but that is outside the scope of this thread.