There is a new version of the livetime counter accumulation routine that has just been placed in the atest.  The old version had a bug that returned the livetime starting 2 seconds before the requested time interval. During longer accumulations, the livetime would resynch with the data after a few seconds  The new version of corrects that bug.
More detail
The data and livetime are accumulated for images and spectra by looping over groups of packets from which the events are unpacked and binned into time and energy bins.  We call the groups packet_bunches.  The error would always occur on the data contained in the first packet_bunch.  That first packet_bunch contained data from 2 seconds before the start of the time_range.  That extended range was used to help determine the data gaps that overlap the start of the time_range.  We use two different sizes of packet_bunches, 1000 packets for 32 bit machines (PC's with 1 processor), and 5000 packets for 64 bit machines (Sparcs and Alphas) and multi-processors.  In 1000 packets there are roughly 250,000 events.  At the highest count rates, when livetime corrections are most important, the first few packet_bunches would have contained the pre time_range data, that would have produced an out of synch livetime, but after 1 second the livetime would have been properly synched.  On the other machines with 5000 packets per bunch, the out of synch condition could have occurred for up to four or more seconds, again depending on the actual rate.
How does this effect your results?:
For spectral accumulations, it would have only affected the first few seconds, mostly just the earliest background. It should be a very small effect.  For short images, the most profound effect is on the coarser grids where the livetime variations are the strongest for events at high count rate.  Depending on the grids you use and the event counting rate, imaging fluxes may change for a few percent to a few tens of percent.  For longer image integrations it will be barely detectable. If you have a paper about to go to press I would not recall it from the journal.  For any short images at high count rate I would reprocess a selected few until you feel comfortable about how your results will change.
So sorry
Richard Schwartz