My personal preference would be keeping the limit as low as possible, even if it means 'Fidelity problem' persists. My anecdotal evidence tells that a couple days ago I had to stop running a full node. The amount of resources it comsumes has become unacceptable to me. The last straw was the power outage that resulted in a corrupted database. The reindexing takes ages, while you can't even comfortably watch Youtube. As the time goes, resources being consumed to do initial-sync only increase. And this is a guaranteed increase, unlike increases in hardware performance. I would personally want optimizations first, and then a limit bump, not the other way around.
Sorry you had a bitcoin database crash. You have learned a basic lesson about running a bitcoin node. A bitcoin node is a database server. It is inappropriate to run a database server on a machine that is not dedicated to running a limited number of servers and has been configured to provide adequate performance for these uses. In the case of data base machines where crash recovery is lengthy, this means it is inappropriate to run it on a machine that lacks reliable (e.g. battery backedup) power and stable software. It is also inappropriate to run a server on a machine that is used to surf the web, or otherwise provide a huge attack surface to the network. It is also inappropriate to run it if the operator doesn't know this, something that you have apparently learned. I learned this a while back pretty much the way you did. As a result I purchased, assembled and configured a dedicated computer to serve as my bitcoin node. This machine resides in my "home office data center".
Most people lack the knowledge and inclination to run their own servers in a reasonable "data center" environment. Until the arrival of SPV clients this was the only alternative for people wishing to use bitcoin. With SPV clients this is no longer necessary, and the truth is that most people would be ill advised to run a full bitcoin node even if they had unlimited network bandwidth and money to spend on a dedicated server. They lack the necessary skills.
Even with a dedicated machine, unless one has excess network bandwidth one may observe degradation of activities on other computers (e.g. watching You-Tube videos) due to interference. With suitable configuration of one's router it is possible to eliminate this interference by limiting the amount of bandwidth used by the bitcoin node. However, doing this requires additional skills, namely knowledge of how to manage and tune computer networks.
Regardless of block size limit, it will be impractical or impossible for the majority of bitcoin users to run full nodes, unless bitcoin regresses to be a hobbyist and/or "science project" operation.
Did you notice that it was anecdotal evidence? Don't take what I wrote literally.
What I tried to show is not how it's bad, but how it's getting worse as time goes on and tx rates rise. Especially the initial-sync part.
And you have only proven my point. The cost of running a full node is increasing. The cost to possess monetary sovereignty is increasing. I'm concerned with that trend. That's it.