This could be a very useful resource for developers. In my opinion, it would be greatly improved by adding reference sequences for the intended / designed strains in each of the mock community. This would enable analysis of errors caused by library preparation, amplification and sequencing which cannot be detected by taxonomy prediction because taxonomy prediction is itself a difficult problem (all current algorithms have quite high error rates for short tags like V4), and for sure taxonomy prediction cannot resolve differences at the level of one or two bases which are potentially biologically informative. Reference sequences would be especially useful in validating denoising algorithms such as MED, DADA2 and UNOISE, but to the best of my knowledge, there are no high-quality sequence databases currently available for any mock community, so collecting the sequences and strains which have not yet been sequences, if any, would be a useful contribution. There is a database for the HMP community of 22 strains, but it contains 113 distinct 16S sequences which is surely too many (it's not clear why; my guess is that it contains all known sequences for the species, not those which are specific to the strains).
I would also note that contaminants are common in the reads. For example, Mock-3 has several reads which are 100% identical to SILVA LN561615.1.1319 (Devosia sp.) and HE798556.395406.396943 (Stenotrophomonas maltophilia D457) respectively, so unexpected taxa are not necessarily errors.
A couple of other mock communities that could be included:
1. Kozich et al. "dual indexing" V4 HMP mock (doi: 10.1128/AEM.01043-13, FASTQ files: http://www.mothur.org/MiSeqDevelopmentData.html). This is yet another HMP V4 run, but is useful because it has a quite different error profile (presumably due to the different library preparation method).
2. The "Extreme" community used for DADA2 validation (doi:10.1038/nmeth.3869, SRA run id SRR2990088). This is useful because it has several species which are >97% similar so cannot be resolved by traditional OTU methods.