The efficiency argument depends a bit on the scope one considers. If you only ever need pipes it makes sense to optimize and learn how to build pipes and your t

The efficiency argument depends a bit on the scope one considers. If you only ever need pipes it makes sense to optimize and learn how to build pipes and your toolkit will reflect that. But if you need a variety of patterns you need to weigh in the inefficiency and cost of switching. The analogy would be a multi-purpose tool versus many special purpose ones (and the knowledge of operating each). This is all theoretical ofcourse, but my sense is that the debate about microservices is colored first of all by the extra complexity of networked computation (which raises the bar for a succesful architecture) and maybe indeed the absence of well defined constraints that would guide people on different possible best practices for segmenting monoliths.