They can but the more GPU cores the more memory bandwidth is required. With the Apple Silicon design using unified memory that memory bandwidth is also shared with the CPU cores. Right now the M1 has about 68 GB/s memory bandwidth which is pretty good for a low power notebook. But according to testing on anandtech.com a single M1 core can almost saturate that total available bandwidth at about 58-62 GB/s. Reference.Your statement is true about CPUs, but not GPUs as I have learned from several sources. GPU cores can actually scale lineary like that or almost. Twice the cores - twice the performance. Regarding ARMv9 Apple already uses many features of it as many have stated so there won't be much performance increase from that. It's ARM who is catching up to Apple with ARMv9.
So any new cores for either the GPU or CPU are going to be heavily dependent on a new memory architecture too. Within whatever budget the next gen ASi chip is defined for there is a limit to scaling. I trust Apple will get it right but it will be interesting to see how they go about it.