To add detail to gkapanthy's answer, the %3N means you have 3 digits of subseconds (milliseconds) while %6N is microseconds. You could use %9N for nanoseconds (dtrace uses this granularity, for example).
We used system strptime at one point, nowadays we have our own implementation which supports a number of common extensions. All the subsecond conversion strings are equivalent if you specify the number of digits.
Note that our internal datastructures still store the data by the second, which means if you have very very large numbers of events in the same second (eg. hundreds of thousands), there can be some performance issues when having to sort them.
EDIT: matteo points out the bit about _time not containing subseconds was incorrect.
