Are you sure you had good joints to begin with?  For QFNs, we have seen very long life in thermal cycle testing.  We haven't done the testing on this thick of a board, however.  But we have also had to do many design tweaks to ensure that we don't have poor perimeter solder joints.  These can have non-wets and insufficient solder that are caused by too much paste on the ground pad.  This relationship could also explain why you have poorer life with larger ground pads.  If your design applies too much paste or too large of paste bricks in the center pad, it will lift up the component and cause opens or non-wets or stretched joints on the perimeter pads.  If your design allows the paste from the ground pad to conjoin with the paste on the pin-pad, it will pull the pin-pad paste into the ground pad, giving either an open or a non-wet.  The non-wetted, starved or stretched joints that test good as-built will have lower reliability.

What is the impact of the thick board?  Is it on the quality of the original joints? (under heated or poor removal of voids?)  Or on the effect of thermal cycling?  (longer dwells needed to reach temperature?)