With so little known right now, where does the mentioned 40% actually come from architecturally? It feels like a hopeful estimation pulled from the air that is being thrown around on the internet. It's an honest question, not making fun of anyone.
It might sound high, but competition-wise, 40% is actually low. Rocketlake is 18% over Skylake, so to achieve 40% you need about the same gain. With variants of Zen 3 coming with amazing technologies like the V-cache Intel should be scared if that's all they achieve.
Another thing is that there's the Intel slide which say it'll be up to 20% faster in single thread. Before all the weird super high Cinebench leak, we thought it was a reasonable expectation. I personally expected/wanted 25% gains per clock with slightly lower clocks. If you believe the Cinebench leak, Golden Cove might achieve 30-35% gain over Sunny Cove!
No one outside of associates close to Intel and the Intel architects know what the gains are. Consider that Skylake is also a 5-year old architecture.
MT performance also has to be considered with Gracemont. Using them for just background tasks will be of minimal benefit. No, with Skylake+ performance and full support for AVX2 they'll be a powerful adder to MT performance.