That sounds like a clever attempt, but also overthinking things. Unfortunately, even if you could connect the parts the way you imagined you couldn't get the intended performance because the low power CPU on a pi board isn't up to the task of managing that array of GPUs. Even if most of the work is done on the GPUs, a CPU is needed to manage what you could call the data flow. What we're calling this "flow" for now is more or less proportional to the work demanded by the GPU(s).
Look up the phrases "gpu bottleneck" or "cpu bottleneck" on google for more detailed information. The idea though, is that a certain balance is needed to get the parts to work.
If you want to mine, you'll need a high end CPU, on a desktop board.