Worst case would be a major disruption just after a difficulty adjustment. Assume after a major global disruption event the global hashrate is only 1/10th of current and can't ramp up for whatever reasons: the next difficulty adjustment would take on average 20 weeks, every block would be mined on average after 100min in the difficulty period after "the event".
The next difficulty adjustment can't drop to 1/10th of current, because adjustments are currently limited to max. 1/4th on the low side and 4x on the high side.
If Bitcoin had to deal with some major disruptive event that severely changes global hashpower some node software fork would be required to cope with this quickly.
And in the opposite case? Where the blackout occurs before the difficulty adjustment.
In the case that a blackout affects the majority (60%) of the nodes and leaves only 30%-40% functional, and this happens right at the moment of the difficulty adjustment, it is highly likely that consensus issues will arise. The remaining 40% could continue producing blocks at lower difficulty, but the chain generated during that time might not be accepted by the 60% of nodes when they come back online.
I still consider the best option to be implementing a checkpoint system, where certain blocks are regarded as global consensus "anchors," for example, those ending in double zero. This could potentially mitigate these situations, creating a fairer environment than simply waiting for things to resolve naturally, which would undoubtedly lead to controversies.
Since we do not know who will be affected, nor when or for how long it will happen, if 60% is interrupted for any reason over a prolonged period, you could always return to the checkpoint established prior to the failure. This would eliminate the need for risk-taking, code modifications, forced adjustments, or entering an unfair situation where bifurcations are debated and lead to significant financial losses. It is not only possible for a catastrophic science-fiction-style scenario to happen but also something akin to the 2010 bug, a vulnerability, or a cyberattack.