This math assumes the fork starts just after adjustment.
You're right, I forgot about that. So would it be best to schedule the fork for the middle of a period to minimize this impact, if that can be done? I think the Doge merged mining hardfork was scheduled for a specific block count, not time.
Yes, that can be done. However, you also can't schedule the fork too close to the end of a retargeting period - otherwise the difficulty won't drop much, even when the hashrate drops significantly. (That's presumably why you said "middle of a period" - this makes sense. At least it should be some compromise.)