The only one of these that is a viable concern is voting, the first two of these concerns are laughable. If i have time i'll outline in more detail, but for potential EOS users who may find this post worrisome, here are some quick corrections.
saying of EOS that "There are no Merkle proofs or any other ways for users to audit any part of the system" is total nonsense, I have the feeling @demphil is regurgitating what Vitalk said in Q&A in China, he was wrong. There absolutely are merkle proofs in EOS and users can verify transactions with only a light client.
Also, your concern in 2 is basically, this scale is too good to be true, but load testing shows these projections to be totally plausible, in certain testing envs it has already gotten up to over 1 million trans per second. I warn people not to listen to heresay when it comes to actual specs, just do a little research. EOS recently reported 10,000 trans per second on a single machine, add parallelism (next year) and that only improves.
Also, regarding concern 3. EOS FEE STRUCTURE, you mentioned that "Even the people who want to use EOS just a handful of times will be needing to buy a lot of coins and sell them all later on". Again, not true. making an EOS account, similar to steem, will allocate you enough coins to functionally have free transactions up to a point, it is designed to enable free usage for someone who wants to use EOS apps. Costs begin to present themselves if you are at the usage scale of say app devs or youtubers who posts dozens of times per day, so you use a lot of bandwidth/storage. In this case you may need to buy some EOS to support your efforts, but at that point you will not be in the group of people who does not "want to spend hard-earned money on an unknown asset in order to use a blockchain", you are invested and it works such that you are willing to go beyond the free use cases to something more.
Great response @cryptokens, appreciate the insight! Would like to hear your thoughts on voting with EOS if you have time.