There's some more discussion about the topic:
and for issues like this
there are some thoughts in Clock synchronization - #8 by Tino (I'm not sure about results... but so far, I'm still convinced that system time alone is not a good choice for timeouts in distributed contexts)