Most of the newer Wu's are checkpointing within 10 minutes.
Some of them are still checkpointing once when first entering RAM then never again.

This one Workunit 49653004
has been running 2d 19h on my 2700X with 3d 17h reported to go and only a checkpoint at the 1st second.
(going to abort it since restarting the client could lose credit on several more WU's)

The WU reports itself as it's 1st time sent to a BOINCer, not a resend.

So are these from the data set before https://www.sidock.si/sidock/forum_thread.php?id=225&postid=1913#1913 where some app routines were dropped?

Will these WU that refuse to checkpoint be exhausted or deprecated soon?

EDIT: (If I look through the WU by time received the answer may emerge. All the ones sent to a Xeon server at 22 Jan 2023, 6:25:11 UTC refuse to checkpoint. Ones sent later to my Intel laptops and AMDs Desktops are almost all checkpointing.)

...has been running 2d 19h on my 2700X with 3d 17h reported to go and only a checkpoint at the 1st second. ...

The algorithm is common for all tasks: If you see, that last checkpoint been too long (for example one hour ago), the best solution - perform restart a BOINC client. Or drop tasks simply. Restart is more better, I think.
Tasks linked to version 2.02 presumably will not face this problem.