Based on the three benchmarks listed at nanoreview, the 12 core Max or Pro with 8 P (performance) cores is about 70% faster for multicore performance, and about the same for single core, and supports more RAM, than the 8 core M2 with 4 P cores.
So it comes down to what you can afford and whether you want to run large, demanding samplesets.
The Air with 24 GB may only load samplesets that fit in 16 GB of memory, which also limits how much CPU you need. If you're happy with the selection of 16 GB or smaller organs (where I'd plan to load 20 bit samples, not 16 bit), then that may be fine. The memory requirements for samplesets are usually on the producer's web site under something like a "Features" tab.
The Pro with 96 GB can load samplesets up to 64 GB, which includes most of them, and has the CPU capacity for all but possibly a few very demanding organs where you might need to compromise on some audio quality settings a bit.
(I'm using Martin's estimate of a Mac's usable memory for Hauptwerk at very roughly 2/3.)