The purity stats with MUMmer don't work well with circular or fragmented genomes, because a perfect alignment is actually split up into multiple different alignments.
The overhanging contigs should be ignored in 5905a879796f63a682cfbebe543861371a591666, but it hasn't been tested yet. Note that this doesn't change things for the calculation of the metagenome coverage.
The purity stats with MUMmer don't work well with circular or fragmented genomes, because a perfect alignment is actually split up into multiple different alignments.
See also:
http://nbviewer.ipython.org/github/inodb/masmvali-publication/blob/master/notebooks/LIMS1884-filtered-pairs-assemblies.ipynb