I do agree that going back to oltp setting would have enabled you to solve your apparent performance problems in a faster manner.
However at the same time it seemingly suggests that when choice of both merge and nl joins are thrown open to the optimizer it is wrongly selecting the merge join option over nl joins at least at some places.
It would be interesting to know what conditions are driving this selection and will it happen even if we provide enough/uptodate statistics information to the optimizer because missing stats may be one of the front running causes.