It's not wrong, because it's not the entire solution. I forgot to add that you also determine fees by checking every 2016 blocks what the average fees per block were. Now miners can't manipulate the fee anymore because they don't know who will find the next block, well unless they unanimously decided to insert fees enough to raise the limit, something really hard to achieve since if just one big miner doesn't agree the rest would be paying him to raise the limit.
It can still be gamed upwards. A miner can mine a block with a transaction paying 1000BTC in fees without relaying the txn. If he mines the next block, he pushes up the average, else he doesnt lose anything.
As has been pointed out, there are disadvantages to both extremes.
The primary cost of mining should be hashing power, not bandwidth/storage costs in a datacenter.
In the ideal situation, the gross mining income with an ASIC should not change regardless of whether it is run on a home connection or a datacenter.
Fees are primarily paid for hashing power to secure the network, not for processing/verification, a single low end smartphone could do that. Once you start requiring a Google datacenter to even enter the mining industry due to gigabyte blocks, you have lost control of your currency to a few, but you will have far higher transaction rates and low cost transactions. What is to prevent this large miner from not charging what Visa does? The barrier to entry has become so high that competing would be a very risky proposition.
Don't get me wrong, decentralisation is expensive, very expensive. Bitcoin in its current form vastly more inefficient and expensive than centralised systems like paypal/Visa due to the need for redundancy in storing/processing/verifying.
The question boils down to what level of mining centralisation are we looking for to drive down fees and increase maximum transaction rates.
We need to find a middle ground which supports as many transactions as possible while ensuring that independent hashing(decentralization) remains viable.