For starters, systems research deals with the real world. The real world is messy. How would you comprehensively represent every interesting workload in an analytical model? How would you derive it from real world applications? It is impossible in the general case (halting problem), and really, really hard to do even in specialized cases. If you think otherwise, the field of worst-case execution time analysis awaits your contribution eagerly (try modeling L1 and L2 instruction and data cache interactions in a multicore CPU).
Thus, in many cases evaluating different interesting workloads empirically is unavoidable. In such cases, 100 pages of setup, methodology, and analysis are a feature, and not necessarily a sign of mediocrity. There is a great danger of overlooking substantial flaws in brief descriptions.
So, yes, mediocrity can lead to inflated sections, but a good PhD committee will not led that slide. In your page cache example, I would absolutely expect to see significant experiments and analysis; I would probably not be convinced by just a few selected benchmarks. Presenting benchmarks well requires many more pages than a succinct proof might.