The answer will be technobabble ...
There is nothing special about four portals.
There is something special about version 0.2.3 of the control software, which became necessary to replace a simple boolean switch (one of two other portals) with a list of destinations. Version 0.2.3 of the software happens to use a newer version of a java library, which fixes a bug in the old version, which affected the tuning of the portal. Basically the tuning compensated for the bug, now the bug is gone and the portal is out of tune.
Of course the researchers tried a rollback to the previous portal configuration and software version 0.1.9. But due to a misconfiguration, the rollback did not undo the library update, which was just a maintenance release with the same major and minor version number.There is nothing special about four portals.
There is something special about the fourth portal, which differed from the previous three. Perhaps all four were handmade prototypes, or four was the first pre-production run, or something like that. The fourth portal produced some sort of static which interfered with portal calibration across the multiverse.
This kind of problem is nothing new for the portal operators on the other world. Instead of discarding an expensive portal, they are able to re-sync their network and to compensate. Unfortunately they didn't know about our experimental network. They have more powerful portals, and also more of them, so their calibration process integrated all four of ours into their network. Anything we send, their (more powerful) receiver will snatch. A few of the things they send, our (less powerful) receivers will snatch. That happens only with their lesser, less well maintained portals, the sort used for low-value cargo.