It 'fixes' the incentives only in cases where nodes throw away data needed for any future validations of the transactions. Whether or not this 'pruning' is a good idea is a matter of reasonable debate.
Incorrect. Regardless of if you prune the data or not, you don't need to access it, so it doesn't impact your working set size.
Turning fully-validating nodes into non-validating nodes is a rather funny definition of 'backwards compatible'.
They continue to validate everything they've always validated. They continue to prevent inflation, prevent double spending, -- they don't validate the new segwit things, but the user with the old node is not using those things themselves, and they know they don't validate them-- so they don't relay or mine them.
UTXO set size is some function of (# users) * (# of addresses holding value per user). As far as privacy is concerned, best practice dictates distributing your value across several addresses. Are we to follow The SegWit Omnibus Changeset with a recommendation for each user to hold all their Bitcoin on a single address? Privacy be damned?
Segwit provides absolutely no pressure to use fewer addresses for your managing your own coins. As I mentioned, it shift fees from the output spending time, to the output creating time-- so it is generally fee neutral for the user except for outputs which are never spent.
Introduction of a new fixed centrally-planned variable?
"1" is also a variable, there is no such thing as a neutral option there.
Preferential incentive for offchain transactions over onchain transactions? Myopic much?
Nothing about segwit is "preferential for offchain"-- if anything it's slightly the opposite.