You're also ignoring the fact that the other statistics are inaccurate. If you're saying the 32 compute units is correct then you have to also agree the 1000MHz maximum frequency is correct, which would mean these compute units are not running at full speed, ergo, these scores are not what they should be.
In the M1 the maximum clock speed of the GPU is 1287MHz, which would mean in the OpenCL benchmark the GPU is only running at 77% of what it is capable of. At 100% it would bring those scores very close to 80,000. Which guess what? That's 4x the M1 !!!
You're also ignoring the scores of the 8c and 16c tests. 8->18,000, 16->38,000. If 8 more cores upped the score by 20,000, wouldn't 24 more cores up it by at least 60,000 yielding a score of at least 78,000?
Something is obviously wrong with Geekbench and its support for these M1 SoCs.