I think the 4MB calculation is too optimistic. This would require that blocks are basically full of Segwit
transactions,
Nope.
They need to be Segwit, but they also need to be signature-heavy. People using Multisig transactions are still pretty rare, and that's the only way to use more than 2MB per block using the Segwit capacity upgrade.
Lightning will bolster this (Lightining involves multi-sig at the blockchain level), but this depends alot on the patterns of usage Lightning attracts. Realistically, alot of BTC will stay permanently hopping around/through Lightning channels, and so the increase in Multisigs may well be a major surge as the Lightning network accumulates it's natural useful quantity of BTC. After that, mined coins will gradually leak in to maintain that balance (whatever it turns out to be), but that doesn't seem to me like a feasible way to fill up an extra 2MB of every block.
After that, other 2nd layer networks (which are planned, but not as well developed as Lightning yet) may well provide the extra. We will see, the 1MB:3MB ratio was apparently chosen very carefully by the Segwit designers, so it's a big test of their engineering analysis.
which is cleary a ridiculous proposition, because a big part of users still wants to use
the "traditional transaction" type and some users still use outdated wallet software, which will never support Segwit.
Nevertheless, I think that Segwit adoption will increase in the long-term and therefore in the future blocks will be bigger than they are now,
but still I expect an average blocksize that doesn´t even get close to 4MB.
I expect that a combination of the lower fees, 2nd layer networks and Bech32 addresses will convince the overwhelming majority to adopt Segwit addresses (Bech32 addresses will be easier to handle, no UPPER CASE lower case distinction, you can almost read them out verbally to someone reliably)