PcoA is only good for visualizing multi dimensional data in 3D or 2D. It is a heuristic and real distances (in your case beta diversity here unweighted unifrac) must sometimes be bend to fit into the 3D space. Good for eyeballing, bad for statistical tests!

You might want to perform a permanova test (http://scikit-bio.org/docs/0.5.1/generated/generated/skbio.stats.distance.permanova.html) to check if there are significant differences of distances between two groups of samples vs. within those groups. If you e.g. group "sick" vs. "healthy". It is suggested to have at least 5 samples per group, thus you have too few.

If you want to check for significantly different abundant taxa between two groups of samples, you could run ancom: http://scikit-bio.org/docs/0.4.2/generated/generated/skbio.stats.composition.ancom.html

But honestly, I don't think that you can find anything statistically sound with only 9 samples :-/

Thank you Stefan for your answer and suggestions.

I share the same concerns, regarding the limited no. of samples, but it is what it is. 

I'll have a look at it. 

AGGS

you could try to put your samples into a larger context: https://visibiome.org/

Thank you.