So you want to build a sVDEV? Look no further, I started a write-up. It’s made a tremendous difference in my use case.
An sVDEV has risks associated with it, but the benefits far outweigh them. A single pool can host datasets of varying types, some of which can reside entirely on SSDs for fast performance while archival-oriented datasets can reside almost entirely on HDDs to save $$$. Yet all of them will benefit from super-fast metadata and the small file performance can also be boosted significantly.
Whether all that makes sense in a video-editing setup with that many disks is a different question. I’m glad your card is 100GbE!!! I imagine you’d want to go 21 2-drive mirrors for the highest IOPS possible. Whether a sVDEV makes a huge difference at that point is a different question.
Lastly, a 21-VDEV pool consisting of mere mirrors will have a higher risk profile than a smaller collection of Z2 clusters. I presume you’re using this mostly for scratch and not archival purposes? If so, all will be well. Otherwise, figure out how to safeguard the data as 21 opportunities for 2 drives to hose the pool is a significantly higher risk than I can live with (I use a Z3 - but my use case is a quasi WORM pool).