You're not addressing the incentives of those who develop new full node tech, which is what really matters in the long run in my opinion. They will likely be incentivized to keep their competetive advantages proprietary and launch in competition with incumbents rather than collaborate. As far as I can tell, the result of the design is that ROI on node R&D is maximized, meaning TPS optimization, which is what I'd prefer from a common computation infrastructure.
Also, ignoring that, lets not forget that centralization of block production at worst only exposes people to what is essentially very complex and ridiculously expensive phishing attacks.
They're a theoretical problem, but easily solved since economic dapps will run their own secondary consensus algorithms in the virtual machine to validate in a way that is tailored to their own security preferences (DPOS is probably gonna be a first choice in many situations). With that in mind Ethereum block validation is ultimately the transaction processing and anti-spam layer, not explicitly a trust layer. For that function casper seems to be the best algorithm invented yet.