That behavior is normal. Or perhaps even abnormally good. xntp predicts the arrival of a packet from its peer. If that prediction is off by 128 milliseconds or more it gets mad and resyncs. When your peer is across the internet, that is guaranteed to happen several times a day. At least, that's what I thought....you must have an awesome isp or something.
I believe error messages until I have a sound reason to doubt them. If your peer is too far away, then your peer is too far away. The closer you are to your peer, the better. And best of all is to have your peer on the same lan.
Your dispersion is 10.01 which I think is great. I wouldn't worry until it gets close to 1000. At 1000 your system clock could be a full second off....I won't tolerate that. But people who sit around tuning ntp until the dispersion drops below one are crazy unless they work in astronomy or something.
Your setup could be improved quite a bit if you want to. You should have a ntp server outside of any firewalls and it should have 6 or 7 peers, not just one. Then your internal boxes should sync up with it.
And the Ultimate Solution.....
HP sold off all of its cool non-computer stuff I don't remember the name of the spin-off company. But they sell HP's old atomic clock. If you get that with the high performance cesium beam tube you will have a clock so accurate that no computer in existence can sync to it. It has a serial port and you can plug it in to your system and configure ntp to use it. That's the most accuracte clock available for sale as a product. Just a suggestion
I don't remember the name of the spin-off company. But they sell HP's old atomic clock. If you get that with the high performance cesium beam tube you will have a clock so accurate that no computer in existence can sync to it. It has a serial port and you can plug it in to your system and configure ntp to use it. That's the most accuracte clock available for sale as a product. Just a suggestion 
