BCH's approach would eventually centralize the network.
Remember that at least one of these blocks is generated every 10 minutes, and more than one might be flying around during a chain split. Remember also that nodes do actually need to iterate over all the transactions in the block to check they are valid when they receive a new block. Proof of stake prevents malicious actors from wasting CPU/DiskIO on full nodes with spam attacks, but the transactions in a block with PoS signature still need to be validated, ideally before the next block arrives.
Originally with the 1MB block, there were quite a lot of full bitcoin nodes running on raspberry PI's under peoples desks in places with really shitty internet. Segwit kinda-sorta actually increased the block size from 1MB to up to 4MB. 16% of Malaysia gets internet slower than 256kbps. Running a bitcoin node today takes up ~20% of a connections total bandwidth in these places. Increase the block size to 20MB, and it will not be possible to run a node in some areas of Malaysia, because 20MB/10min is too fast for the connection.
At 20MB of transactions per 10 minutes it's also possible that the diskIO on a gen-1 Raspberry PI using a cheap SD card might not be enough to scan and validate every transaction in an incoming block. Remember that a node might have to scan very far back in the blockchain to find the last time an unspent output was interacted with. I'm not sure how big a block has to be before that IO overhead starts eliminating entry-level hardware.
But you don't actually get much for increasing to 20MB blocks. Another poster said that bcash going from 1MB to 8MB blocks increased the throughput from 7tx/s to 62tx/s. Lets take that as gospel, assuming that's 1MB blocks vs 8MB blocks, and extrapolate like madmen. To get to that posters desired 7,000tx/s "Visa scale", we'd need blocks of around...1GB.
That's 13.3mbps 24/7. Now most of Australia and the USA can't host a full node, and that's before we start talking not just about disk IO but even hard drive space. If new blocks are being created at the rate of 1 Gigabyte per 10 minutes, you'll need to provision 4 terabytes of new storage every month to compete with Visa.
Look, short version is this: scaling by increasing the block size eventually ends in performance requirements that force datacenter scale hardware requirements. If bcash ever reaches that point, it will have lost the decentralized, censorship resistant attributes that make cryptocurrencies valuable in the first. If your cryptocurrency isn't decentralized, you might as well use a bank. There's better regulation and much less uncertainty in being a customer of a bank.
So anyway, the Lightning network developers think that LN provides a better scaling solution. Maybe they're right. If they're not, someone will try something else. Maybe there's no solution that can scale to Visa scale, we just don't know. But bigger blocks is definitely not that solution.