That is a tough question to answer. I toggle between the two approaches myself. Some datasets I've analyzed work much better with deblur, others with DADA2. Generally speaking, I think it has to do with how the quality control and denoising occur.
TLDR; See this post.
-Mike