Perhaps Blockstream never wanted SegWit adoption. That 95% activation level seems to have been designed to fail from the get-go.

BIP9 changed to a new quorum sensing approach that is MUCH less vulnerable to false triggering, so 95% under it is more like 99.9% under the old approach.  But we saw no reason to lower the criteria:  basically when it activates the 95% will have to be willing to potentially orphan the blocks of the 5% that remain if they happen to mine invalid blocks.   If there is some reason when the users of Bitcoin would rather have it activate at 90%  (e.g. lets just imagine some altcoin publicly raised money to block an important improvement to Bitcoin) then even with the 95% rule the network could choose to activate it at 90% just by orphaning the blocks of the non-supporters until 95%+ of the remaining blocks signaled activation.

Miners could simply use software that does not recognise segwit rules (such as earlier versions of Bitcoin Core) to mine blocks on top of a chain that has activated segwit. This would be a hard-fork as far as segwit-aware software is concerned, and those blocks would consequently be ignored by Bitcoin users using segwit-aware validating nodes. If there are sufficiently many users using segwit nodes, such a hard-fork would be no more effective than introducing a new alt coin.