Nice to see Jeff finally got some success running GPUs on the pi. Though, as the conclusion kind of points to, it's not very practical today. To be practical I guess we'd need to see faster bus speeds, newer GPU support AND probably, a more practical way put the setup together (I'd love the see a Ghost Canyon[0] style setup) which would make building a high performance ARM system at home practical.
[0]: https://www.pcmag.com/reviews/intel-nuc-9-extreme-kit-ghost-...
Minisforum is also selling a new option with an external GPU bracket/dock[1]; honestly the same thing can be replicated with a Pi Compute Module, though nobody's making a nice enclosure for it yet[2].
I would like to re-test things once the CM5 comes out, but as you state, more PCIe bandwidth is better. Would love to have at least Gen 3 x4, that way you could also get more bandwidth between other devices too, like NVMe + GPU + 2.5 Gbps NIC.
[1]: https://www.tomshardware.com/news/minisforum-b550-amd-ryzen-...