No, the metric is only used in choosing routes (the metric reflects the cost of the route).
As Jaap says the variation you're seeing is more likely just variation in system load and other (non-networking) factors. Getting reproducible test results can be a challenge - you really need to control what's going on on the system at all times (for example I've seen overnight performance stress tests fail when the system went to check for OS updates at 2am: the extra CPU going to this other task was enough to push the system over the edge into failure).