The 95% miner signaling is important to prevent two Bitcoin forks, such as what happened with Ethereum HF and Ethereum Classic.
Bitcoin has a very slow difficulty re-targeting algorithm. A fork that has just 95% miner support will initially (for 2016 blocks) be 5% slower (an average block every 10 minutes and 30 seconds). The transaction capacity of the new Bitcoin protocol is reduced only 5%.
However the chain with 5% if the hashing power not only has a 20x capacity reduction, but confirms transactions in 20x more time. So the mempool will grow 400 times. It must be noted that fees increased 10x from the moment blocks were half full, to the moment blocks became saturated. I'm sure no Bitcoin (pre-fork) user will be willing to pay 100x times the transaction fees to use such a slow and insecure network.
So a 6-block confirmation will take 20 hours in the original chain and the original chain will be in this almost useless slow state for an average of 2016 blocks, or 280 days.
If the original blockchain hard-forks to re-adjust the difficulty, then it will just represent an alt-coin having 5% of Bitcoin community, and it can't affect Bitcoin (the segwit2mb fork).