There's a very long standing Sonar bug (several versions old) that seems to affect loop length when loop recording so that either the last few milliseconds of each loop are missing or the loops are different lengths.
For whatever reason most systems seem immune to this bug. Mine was immune when using a Delta44 - it appeared when I switched to a (Cakewalk badged) UA-101 interface.
In my case the work around is to use WASAPI drivers (which also gives me significantly lower latencies than using ASIO with this interface), which would seem to indicate it's probably a driver-related issue one way or another.
I reported it as a bug well over a year ago, to get the response that the report had been "passed to developers". None of the updates scince seem to have sorted it out though :-(