There's an interesting sociology and economics of science and academia which could use more attention and ... research. This includes why and how specialisation and focus on narrowly-scoped rather than synoptic, integrative, systemic, or "generalist" study, scholarship, and instruction emerge. The problems are not especially new or specific to our time. Yes, there were in the past scientific generalists who studied or contributed to a large number of fields, but there've also always been those who guarded their own domains, not necessarily scientific, against such approaches. Jealousies between science and religion, law, medicine, and crafts or (technical) arts, for example.
I think that trend is more generally human and speaks to tribalistic tendencies. Those themselves emerge for numerous and complex reasons. Innate complexities of fields, difficulty in judging or assessing new contributions, an inherent conservativism focused on what is rather than what might be, of preserving established order against change, and the like, seem major components.
I'll note that in the computer and technical field, as I age and hopefully experience, I increasingly find myself in the more conservative camp myself, just for the record.
On compensation and reward: it's been observed in education where much has been made of metrics focusing on individual teachers and students that those indicators seem to hold very little validity. The strongest evidence of this is for teachers working at multiple institutions, either simultaneously or transferring within a given year. Looking at those datasets, there's no correlation between performance between the two.
I'd argue similarly that attemtping to reward or penalise researchers based on individual performance or papers is ... a false metricism. That true assessments of value are difficult (and that most metrics used --- papers published and patents issued, for example) are meaningless. There's little agreement even on what, say, the ten most significant inventions of all time have been. (And were any of those developed in the past 100 years?)
Cultivating a research cadre, or sets of cadres, ensuring that they are supported, testing where appropriate against measurable criteria, but allowing for a high level of fuzzing and uncertainty in those metrics, and looking more to aggregate performance of teams, institutions, disciplines, and the like, might be a better approach. Markets work poorly here. Institutions and patronage in its various forms (including national support) seem far better suited.
And yes, revisiting and testing of recent and older results should be a constant part of that process.