This is fixed now.
The configuration for the "Sapling 2023h2" CT log was incorrect, which resulted in all our SCTs going to other test logs. Two of those other test logs became overloaded, causing slow staging finalizes and intermittent failures.
I've correctly configured Sapling 2023h2, and everything should be better now. This would have started at 2023-07-15T00:00Z when the Sapling 2023h1 log shard ended.