A strong malleability fix _requires_ segregation of signatures.
No, none of the alleged benefits of SegWit requires segregation of signatures:
* Malleability could be fixed by just skipping the signatures (the same data that SegWit would segregate) when computing the txid.
* GREATER bandwidth savings could be obtained by providing API/RPC calls that simple clients can use to fetch the data that they need from full nodes, sans the signatures etc. This enhancement would not even be a soft fork, and would let simple clients save bandwidth even when fetching legacy (non-SegWit) blocks and transactions.
* Pruning signature data from old transactions can be done the same way.
* Increasing the network capacity can be achieved by increasing the block size limit, of course.
It's size sets a hard lower bound on the amount of resources to run a node. The fact that the size limit doesn't reflect the true cost has been a long term concern, and it's one of the biggest issues raised with respect to blocksize limits
Biggest issue of this week, perhaps?
Surely you know that the non-mining relay nodes invalidate the few security guarantees that the protocol can offer. Simple clients should not connect to them, but to miners (or relay nodes that are know to be run by miners). It makes no sense to twist the protocol inside out in order to to meet CONJECTURAL needs of those nodes.
The only cost that really matters is the marginal cost for a miner to add another transaction to his candidate block. That is the cost that the transaction fees have to cover. The magnitude of that cost is one of the great mysteries of bitcoin, extensively discussed but never estimated. But it seems to be very small (at least for competent miners) and is probably dependent only on the total size of the transaction. But anyway the developers have no business worrying about that cost: the fees are payment for the miners, it should be the miners who decide how much to charge, and for what.