This concept of course not going to amount to anything more than a hobbyist or prototyping/POC use for quite a while purely from a cost perspective. You can hire a master C programmer and pay him $100 an hour for a year, or an advanced JS programmer and pay him $40. Well after launch, you may have saved $150k, but your hardware costs you $50 extra per unit to produce, requires more power to run, and has lower performance and timing precision (not real time). Your break even point is now 3,000 units. This is a non-starter. Other engineering, design and marketing costs will likely land you in the red if you are only producing 3,000 units. Perhaps in some niche markets this concept might be viable (like $500k plug in medical devices), but I doubt it.

There is no way to really getting around learning hardware. Who is going to write the hardware interface drivers or driver/JavaScript bridge.

Power consumption will rule out the possibility of many portable devices, loss of performance will rule out many options as well as will latency caused by the garbage collected (non real time) nature.