Hi @blau,
dada2 typically produces fewer features than OTU clustering, as the error correction and chimera checking steps employed by dada2 will remove spurious features.
However, dada2 is also more sensitive than OTU picking in that it resolves exact sequence variants (100% OTUs) that would be collapsed together into fewer OTUs during OTU clustering at, say, 97% identity. So under some circumstances (e.g., where sequencing error or chimera occurrence are low), it is conceivably possible for OTU clustering to detect fewer unique features that dada2.
Additionally, the number of features detected by QIIME1 also depends on additional quality filtering steps, e.g., the filters applied during split_libraries_fastq.py and if chimera checking is used. So your results may not reflect those reported by others (if they do not apply the same filtering methods)
@gregcaporaso and @benjjneb may have additional thoughts to add.
I hope that helps!