I think this is more a testament to the quality of Adobe's GPU pipeline (or lack thereof). M1's GPU is surprisingly capable for a 10W part, and more importantly, it supports low-latency CPU/GPU communication and opt-in memory page sharing. The later is especially interesting for creative software, as CPU and GPU can work on the same data structure, eliminating the need for copies or GPU buffer management. Knowing Adobe however, it's probably most likely that they are still using OpenCL, so the lack of performance is not surprising.

P.S. Or maybe the GPU test is simply memory bandwidth limited...