The industry is starting to recognize this as a systematic problem. For reasons currently unknown, BGAs assembled to ENIG-plated boards that are exposed to a second reflow operation periodically experience a severe weakness in the solder joint or, in more extreme cases, dewetting. If you could send me some images of this defect, I can give you some feedback if this is indeed your affliction.

I have seen this type of problem before. Although you may or may not be mechanically stressing it, it is likely it is being stressed during handling, packaging, shipping or when your customer is handling it. Unless it is a backplane and extremely rigid this would likely be the culprit. The solution is handling and storing the boards on a flat surface. ESD safe plastic trays are the best way, as the board is fully supported on a flat surface. The operators are handling the tray, not the board, the board will never see stress. You and your customer need to use the same method for handling. The prior industry method before trays and tray carts was carts or racks with slotted side panels and the boards would hang from two edges and sag in the middle and frequently fracture solder joints on BGA’s and or circuits. These older rack methods are still around, but the industry is changing. You can get trays and tray carts from a variety of companies, including mine.

I hope this solves the problem for you.

As a field and service rep in my past life, I have experienced the same types of situations. If either or both of the customer's facilities are close enough to your facility to allow you to visit them during the assembly stage where the failure is most likely to occur, I suggest you schedule a trip to observe the assembly procedure/methods they use. If both customer's experience the same problem having the same or similar design, then it must be in the assembly method used or the tooling they use that is causing the problem. If the problem only existed with one customer, then you could deduce that the client probably had some assembly issues especially so if other clients used the same methods and design without problems.

I hope this helps.

You may want to try a step up stencil in the BGA area; step up to 8 mil thick.
This may help with large BGA's

In my opinion....40% of the task is materials/solderability....20% is mechanics (alignment/placement/pcb+component handling)....40% is thermal profile.

I would look at making sure your chemistry is correct....and also your thermal profiling. Are you warping the pcbs during reflow? If so, a lot of stress will be induced in your assembly...which will lead to subsequent failures. You may need to slow the preheating rate down, allowing heat to evenly perculate through the whole structure, to reduce the warping. Also the reflow temperature + soak times need to be sufficient to allow the bonding to occur...your solder manufacturer will have guidelines to follow.

Only my thoughts...