First of all you will be notified when you try to publish that an experiment is running with the option to either "Cancel" or "Publish". If published, the A/B test settings will always take precedence. This means that;
If A/B test is run on 100% of the traffic, the change in main site configuration will not show to anyone.
If A/B test is run on < 100 % of the traffic, the change in main site configuration will show show for the part of the traffic that is not included in the A/B test.
