As of now, in Feature Experimentation, if a variant performs (very) bad, there is no way to deactivate it or set its behavior back to baseline without needing to create a new rule.
This of course slows down experimentation speed.
I do understand that the results of that variant are not usable after setting a variant's behavior back to baseline behavior. However, that is not the issue. The goal is to simply be able to continue letting the test run while disabling a bad-performing variant.