[atlas]internal timing constraints?

Dear Atlas folks, do you have a rough estimate or data about the internal timing of the probes, like what sort of "delay" is added by the probe itself when measuring e.g. ping RTT? The background is that I get different readings or ping RTTs from the probe, compared to the "same" thing done from a regular PC running Linux and sitting on the very same subnet, connected to the very same (small) switch. The absolute difference I see is small (sub-millisecond), but it seems to be systematic. The ping target is a "local" instance of i.root, btw. Thanks, Wilfried.