Several ways to do this.
Given your use case, I would open the VSI and change the Think Time for each desired response to 3000 millis. If you had a really high number of responses, you could edit the VSI file and perform a mass find/replace -- but if you do that, you are taking on a risk of causing irreparable damage to the VSI.
You could also access your service in the Portal and change the Think Scale to 60%. This change will reduce all of the response think times to 60% of their value as recorded in the VSI. So, if a response is 5 seconds, you are reducing it to 60% of 5 seconds which is 3 seconds. The danger here is that the service is redeployed, the response time is set back to the original time slice (e.g., 100% or 5 seconds).
You could also add custom logic to reduce the think time. One way is via a Scriptable DPH in the Responder step. This approach is less visible and rebuilding the VSM could overlay this code. You could get a little more fancy and add a config property off of which you could change the response time.
If your change to 3 seconds is a permanent one, I would recommend changing the response time in the VSI to prevent the unwanted side effect of someone accidentally changing the virtual service model behavior.