I think I've found a workaround, but I'm not sure why this has to be done this way.

Each log even will only have 1 matching lookup out of the 5. So instead of having a multicast event that feeds that data into each of the 5 lookups, I just feed the data into the 1st lookup. Any non-matching rows from the first lookup are passed on to the 2nd lookup, then the 3rd and so forth.

This works if I don't multicast, and I've successfully executed a test run of 2,400,000 records without any issues.

But I'm not sure why the multicast would cause this crash, has anyone experienced this before?

EDIT: None of the other changes matter to stability if I flow the data this way. It works with fast load, full cache, no cache, etc... so it does seem to be the multicast.