Cluster cutoff

I analyse sequences from Ion Torrent.
When I use the cluster command (av method), the cutoff change systematically.
Yet I have made ​​an effort to trimming with qaverage=25 and diffs=2 for pre cluster. Given that these are sequences from Ion Torrent, I know that the error rate is high.
However when I use the nn method under cluster command, the cutoff does not change.
My question is why? and even if the nn method is time consuming, it is as “good” as the av method?

Many thanks!!

My question is why?

http://www.mothur.org/wiki/Frequently_asked_questions#Why_does_the_cutoff_change_when_I_cluster_with_average_neighbor.3F

and even if the nn method is time consuming, it is as “good” as the av method?

No. http://www.ncbi.nlm.nih.gov/pubmed/21421784