You could have a timelocked transaction that has a zero value input (and other non-zero inputs). If the SF happened, that transaction would become unspendable.
The keys to the outputs may be lost or the co-signer may refuse to cooperate.
There seems to be some objections to long term timelocked transactions.
If someone asked me about it, I would recommend that any timelocked transactions should very carefully make sure that they use forms that are popular.
I think the fairest rule would be that any change which makes some transactions invalid should be opt-in and only apply to new transaction version numbers.
If you create a timelocked transactions with an undefined version number, then you have little to complain about.
If the version number is defined and in-use, then transactions should not suddenly lose validity.
A refusal to commit to that makes long term locktime use much more risky.