That is certainly interesting. I have never seen a DH+ problem that required reloading software. You may be able to restart communication by manipulating the message control bits in ladder logic. Make sure the message rung is not enabled and then unlatch the error bit, clear the error value, etc. I have had to do similar things with C-Net message errors in a PLC5.
You may also want to monitor the DH+ status words on all four machines to compare the health of the networks. See if you are getting a bunch of retries or timeouts on the two that are troublesome. The channel must have a diagnostic file assigned before you can monitor those registers.
The cabling may look perfect and still be bruised or picking up electrical interference. Double check terminating resistors, node numbers (including the PC KT card) etc...
Also, be sure you are not over filling the message queue in the SLC. The true nitty gritty details of how DH+ works are hard to find, but there are papers out there on ab.com. There are big differences between PLC5 and SLC504. It is possible to overload the message queue. Make sure you interlock messages so that you don't trigger too many of them at once. Also, having a programing terminal connected via DH+ will hog bandwidth too and could exacerbate the problem.