Pat, if I understand you well, I should get rsq_1 < rsq_2 < rsq_3 (and not the other way around as I wrote above). But I get rsq_1 < rsq_3 < rsq_2.
This is very briefly discussed in an older forum post (https://forum.mothur.org/t/pcoa-r-squared-error/1573) but in my case not so many distances are close to 1.
According to which criterion shall I select which OTUs (out of the 9083) to plot on the biplot? The length? The p-value (and if so, 1, 2 or 3)?