As per bug 1366217 we occasionally see timestamp inconsistencies when gathering GC statistics in release builds. There are assertions around these but we we don't see these fail during testing. We should add some telemetry for how often this occurs in practice and on which platforms.
We see this in nsTimer as well periodically as well. It is always on the windows platform. There is some code in TimeStamp on windows to switch between hardware counters and a less accurate method if we detect the hardware does not work. Perhaps that is what you are seeing here as well. +mayhemer who knows about the TimeStamp impl.
You need to log in before you can comment on or make changes to this bug.