Your figures for DeepPrime suggest that your GPU does not contribute much, if used at all.
Compared to your figures, the GPU in the M1 MBA seems to be fairly helpful. DxO has not published anything yet that is optimised for Apple Silicon SOCs, but the apps run well on the new chips and Rosetta seems to be up to the task too.
Meanwhile, I tested develop times depending on how parallel processing is set and on the number of pictures processed. Here are the figures for the total processing times for 1, 2, 3 etc. parallel images and the No Correction preset.
05 images: 13, 08, 07, 08, 09, 07, 07, 07, 08, 07, 07, 07
15 images: 41, 24, 21, 19, 19, 22, 22, 22, 22, 22, 19, 25
As we can see, times don’t change much above a setting of 3 and I’d not venture into saying which setting above 3 gives the best results in all possible cases. Default on my iMac was a setting of 6.
Maybe the sweet spot is at 4, taking into account this post.