I totally agree with what @gregcaporaso suggests.
We recently analyzed our new dataset using both QIIME 1.9.1 and QIIME 2. What we found was that QIIME2, either denoising by dada2 or deblur, performed much better than QIIME1.9.1 in terms of the accuracy of taxonomic assignment, as evidenced by the far superior estimation of mock compostion by the QIIME2 pipeline. The beta-diversity, however, showed no different results. We saw the similar clustering of samples and both pipelines gave the same conclusion.
Now in QIIME2 2017.11, you can easily run both 1.9 and 2 pipepline and compare your results.