Yes that is biggest thing that am trying to find out from the last 2 days, But no clues on it but as you told some thing is wrong. was using one forloop(to read the old qvds and convert them into new once) before the exit script but i dont think it will cause any problem as it is well tested in dev and worked fine.
If I look on the various questions from you to this topic I think it would be useful to split your tasks into several tasks and further to slice your rather big qvd's to smaller ones maybe on year or yearmonth level. Of course you will add some overhead but it will be a lot easier to handle these smaller tasks / qvd's and especially a potential trouble-shooting is a lot faster.
Additionally I suggest to consider if (more) incremental load-approaches could be applied. They are not only useful by loading rawdata else within a second/third data-layer, too.