Core ML models can only accept RGB or BGR images as inputs (the alpha channel is ignored). If your model wants a YCbCr image, you'll need to put the image into an MLMultiArray instead. You'll also need to change the input of the Core ML model to be a multi-array, not an image. Alternatively, you could add some layers at the start of the model that convert RGB back to YCbCr, but that seems like more trouble than it's worth.

In general, with MLComputeUnitsAll CoreML will try to use the ANE first, but there is no guarantee that it actually will choose the ANE. For more details, see https://github.com/hollance/neural-engine

I don't know what the answer is, but I would try running the model in CPU-only or GPU mode to see if it happens there too. Perhaps it's an issue with the ANE. In any case, these developer forums are pretty deserted. If this is an important issue for your business (and it sounds like it is), you should use one of your Tech Support Incidents or contact a developer evangelist from Apple.

It takes as long as it takes. Using multiple dispatch queues is unlikely to help, since you're limited by how fast the hardware can go. If you want a faster body pose detection method, you will have to train your own model.

Does your model use a NonMaximumSuppression layer (or as part of a pipeline)?