I've just checked the most recent run (2450, same problem as 2444).
I've looked at the QRT and MPS signals in the counting house. As they come from the source, the QRT transition occurs ~50 ns before the leading edge of MPS. After delays these are sent to the hall with QRT transition ~55 us after leading edge of MPS. So there seems to be no problem upstairs.
This is odd. A problem downstairs could account for the QRT not lining up with T9. But there seems to be no way for a problem downstairs to cause the sporadic early QRT seen in runs 1953 and 2385. Maybe there are two separate problems.
A copy of this log entry has been emailed to: bogdanw@jlab.org,cates@virginia.edu