I doubt many people would have done that, since Mathematica's Linux-ARM port is only intended (and licensed) to run on Raspberry Pi boards.
While not technically impossible, this kind of setup would make sense only for very specific workloads that can take advantage of multiple kernel parallelism (using ParallelEvaluate
and friends).
Note for example that each board has 8 CPU cores, but only 2 GB of RAM to be shared among all kernels running on it, which is likely to be a limiting factor.