I agree that too much decentralization is wasteful and can erode performance. The question I would like to ask is: What is the goal of security through decentralization? As BM stated, there are many variables that factor into security and security means different things to different people, but IMO the most important part of this aspect of security is simple: protect user assets from being stolen. Whether through direct theft of assets, inflation of the native asset (bts), manipulation of price feeds, ect. Most all attack vectors revolve around stealing user assets or making their assets worthless. I would argue that much of this has been solved through cryptography and giving witnesses as little privileges as possible and giving delegates delayed privileges to give users a chance to vote out bad delegates.
It would be nice to see, if possible, what percentage of votes cast are via unique proxy (slate) to determine true decentralization of the network. If, for example, >50% of votes are cast via 17 unique proxies, then It may not be advantageous to have more than 17 geographically distributed witnesses at current adoption levels. As long as the network is sufficiently decentralized to protect user assets, then it works, and perception should not be a concern for something that works just as efficiently or more efficiently than its competitors.