twist: I had correctly identified it as a "multiple copies of supposedly-unique thing running concurrently" issue, then saw logs that appeared to confirm that and latched onto them, but I forgot to filter the logs to just production. production was doing one per minute as it should, the other 37 are the task scheduler in various dev/testing environments firing off their own instances of the same task

it was actually just several copies of the production thing enqueued in separate minutes, then getting backed up in a queue before the workers reappeared and picked up all of them at once

tl;dr: complex async system debugging is hard and i should go back to bed