The user was running responder with Cisco ASR1k with about 900 nodes. CPU usage was 100% and the ASR1k was dropping about half the packets.
With IP SLA responder, the server is running at mere 2% CPU load while doing lots of other stuff, and you can see from the graphs that the results are better as well. Jitter may seem worse at first glance, but if you look carefully high and low values are same, they just appear more often, because you have more jitter samples, due to non-congested responder. The improvement can be observed at the right end of the graphs, just before 12:00.