I doubt that the server call failure is the direct cause of the time change, but it was a failure to look into to see if it led anywhere interesting and its failure that others see. In the case of your time jump, it seems more likely that the problem lies somewhere with the low level time keeping code, below the routine now(). Another interesting factoid is that the RTC update interval is 60 secs which is similar to the amount of time it takes to jump after its been set.
The symptoms with NTP and manual set are consistent. The system is to the current time when the NTP sync or manual set happens and then jumps ahead within a minute or 2. Out of curiosity, with NTP off, once the time jumps to the future (i.e. jumps ahead by ~49161652s) does the time then stay relative to this jump or does it jump again at some point?
At this point I think Ray needs to weigh in.