In order to determine a user's position, a GPS receiver must know the position of the satellites transmitting the ranging signals to the user. Included in the ranging signal received by GPS receivers is a navigation message that includes, among other things, information regarding the satellites position, timing, health status and atmosphere correction information.
Each satellite sends a new ephemeris to the user for every two hour interval. Agencies like the IGS, NGS and others record this information from their large continuously operating networks. This allows users to post process data as well as analyse the orbit data.
Normally, the the ephemeris epoch occurs every 2 hours on the hour. So as an example you would expect for a single satellite to receive a new ephemeris for 00:00, 02:00, 04:00, ... , 22:00. However, some ephemeris have an ephemeris time of 2 hours minus 16 seconds.
At first sight, it seemed like it could be a problem related to a particular type of receiver or firmware. Additionally, the offset between GPS time and Coordinated Universal Time (UTC) time is 16 seconds. So this reason seemed plausible. After further analysis we found that the occurrence of this phenomenon was much more common then first expected.
After some googleing, I was able to find an explanation for this via the UNAVCO TEQC email list. The issue is caused by the resolution of the messages which is 1 LSB which is equal to 16 seconds. So it was only a coincidence that this matches to the current leap second offset!
It is also possible for other epoch offsets to occur. I've plotted the occurrence of the offsets in the epochs for various offset values from the normal 2 hour interval for the year 2011. The data was sent to Lou Estey by the GPS Operations/Payload Analysis group. The result of their analysis is shown below:
So in the end, the non-standard ephemeris epochs cause no harm, are valid to be used for positioning and are just a small quirk of the navigation message format.