In general, Noda Time has some issues with converting instants to local dates at points close to (or at) the end of time. See e.g. issue 197 and issue 113. This appears to be another occurrence of the same kind of problem.
I'd suggest filing a bug. One suggestion, though: check whether you see the same results with Noda Time 1.0: the fix for issue 197 (in 1.1.0) could conceivably have caused this as a regression.