Oops; I had forgotten about the need to mine 2016 blocks at current difficulty before allowing it to quadruple (and I thought it could at most double). So correcting for both errors, the 10 hours becomes 10000 hours, or well over a year. Throw in more realistic quantum cycle times, constant factor overheads in Grover's algorithm, and quantum error correction slowdowns, and you're looking at many years...
I don't think the 2016 blocks and difficulty adjustments matter for trying to rewrite the history as the attacker will just mimic the existing history. The amount of hashes you calculated would still stand the same to beat the cumulative difficulty of the existing chain. But really, all they need to do is rewrite recent history to perform double spends at will, and the developer checkpoints will prevent very deep history rewriting.
But this line of arguing is pretty pedantic if they can just steal all unprotected funds and funds as they are spent from scripts. (anonymint is very good at sending discussions off course.)