A constant stable performance drag is going to show up in your load tests, capacity planning, etc. rather than in surprising intermittent degradations.

Also our log collector agents run in containers like everything else, so there is some amount of resource isolation (not perfect of course).