Figuring out how to scale-out non-trivial problems is difficult. Many of these scale-out solutions are going to be worse than local processing, and many won't add meaningful value at any scale. However, there's real value in approaching difficult problems, and there's real value in "the fastest way to do X in a cluster" even if it's not "the fastest way to do X". That's because scale-out is a real requirement to do a lot of interesting things, and because if we can find better ways to scale out we can start to compete with single-machine solutions (and "custom cluster" and "supercomputer" solutions) then we have made real progress towards something that's faster at all scales. It's a valuable research goal.
Now, it would be great if the database community (and the systems research community at large) was more explicit about that. They should set the standard that authors need to clearly differentiate between "improved on result X under constraints Y" and "moved the state-of-the-art forward under all constraints".
It's also good to read systems research skeptically, and think about whether you really need to pay the cost of scale-out. For companies who's output isn't primarily "research", it makes a lot of sense to constantly look for "the cheapest way to do X", not "the cheapest way to do X in a cluster".