In which case it sounds like you are updating so many documents that you're hitting all your segments causing them to need restoring.
FWIW this logic hasn't really changed since version 1.4, although the context in which it's used is of course radically different today. I can see some scope for improvements in this area that weren't applicable in the past and opened https://github.com/elastic/elasticsearch/issues/55142 (see also https://issues.apache.org/jira/browse/LUCENE-9324).
I haven't previously heard of anyone else with similar issues, corroborating what I said earlier about this being a strange way to operate a cluster.