Alright, pretty sure at this point it’s related to the “Inline SNMP” polling APC UPS units. CPU usage had remained far more stable after disabling that folder, but would still spike after an hour or two (or sometimes almost immediately) following a restart. Realized I had two APC units not in that folder.
I’ve reenabled the folder, and have added a rule to force the classic SNMP backend for all UPS units. Have been good for the last 18 hours or so. Restarted the site several times, seeing normal CPU usage. Disabled the rule, applied, and within a minute, seeing high CPU.
Will spend a bit of time trying to narrow it down to see if the behavior is limited to APC NMC v2 or v3 and then file some bug reports.