Basically an extra 2000 txs is possible for an extra 3mb of data that requires hard drive storage. Blocksize of 4mb, no funny codings, possible 10,000 txs.
I do recall Core developers, supporters, complaining about storage cost blah blah bleh.
ptting it short
2009-2017..
with bitcoin 1mb. if EVERY tx was a lean 223byte tx, we 'could' have had blocks with 4500tx already 2009-2017..
4500tx:1mb
but the average tx is more like 400bytes so its more like 2500tx a block... bitcoin has topped off at this average for the last 8 years..
now considering a 1mb base segwit 3mb witness cludge
if every tx was lean AND every tx was segwit then a possibility of ~9000tx due to the witness area. but there would still be spare 'weight' buffer below 4mb weight.
best case of average tx size but 100% segwit utility is 4500tx for ~2.1mb...
but even that would be hard to achieve due to spammers sticking to native tx's for malicious purposes so dont even expect 4500tx:2.1mb .. just like many people gave up expecting 7tx/s for bitcoin 2009-2017
the reason for the 4mb weight is not to allow more segwit tx in.. thats not how the ratios work. its to allow future features like CT to bloat up a tx by appending even further data to a tx to make it confidential
leading to at best 4500tx:4mb (if 100% segwit keypair funded tx in the block)
...
now imagine we didnt have the 2 merkle cludge of segwit. and it was just a neat clean 1 merkle 4mb weight where native and segwit can be side by side, no filtering/stripping..
then expect 9k-10,000tx
or expect around 4500tx if things like confidential transactions bloated the block space
the issue with not having the segwit cludgy 2 merkle strippable feature is:
1. core need the 2merkle to help create their tier network and bring them to the top as the 'real full node' dictators
2. core need the 2merkle to simplify soft concensus bypass(trojaning) in new key types and features in on a whim
the parts about LN needing it is not 100% true but the blockstream teams version want to invent new keypairs so need the segwit bypass to slide in their new features that are set to utilise segwits consensus bypass.
yet LN can function just using current features/multsig addresses.
...
im meandering.
the short and curlies of it is if they made a hf where the cludgy code was removed and the full weight was the main block. (1merkle) then it would allow more real capacity into the blocks. cores domineering ability to slide in their features by ruining their tier network would suffer too. but on the plus side bring the community to a level playing field of full blockspace utility without hopes of needing to change to new keypairs, with out the cludge, without the bad math manipulation.
theres alot of stuff most are not talking about because it goes over peoples heads. but blockstream 'need' segwit.
and even this barry silbert stuff is just more drama to sway people into accepting the cludgy code sooner.