There are technically two issues. One is the liquid metal escaping.
Two, which I think is more important, is the bonding (of the silver shim with the heatsink).
On a CPU only workload (cinebench) I get 43/44W consistently with multiple versions of the paste (my Thermal grease, the original liquid metal, and the Arctic Silver).
When I am playing a game with a GPU+CPU load, the bigger active area (of the GPU+CPU) caused the hotspot temperature to drop.
Furthermore, as @PSierra117 has demonstrated, replacing the shijm and modding the heatsink allows for substantially higher thermal transfer, of 62W in cinebench.
This is near 50% higher than stock or Heatsink Compound or Arctic Silver, and still 10% higher than the seemingly PTM results.
Not surprised. GIven that the fan blows on the separate GPU and CPU heatsink, but only one is doing the work.
This is good for expansion shell users, who would otherwise want to believe they are losing out significantly on the fans’ horsepower. Not really.
I do get higher than 14500 (probably near 15000 territory), 7840HS. 3992MHz, sometimes 4017.
But it’s still only 42Watts.
I kept saying, “don’t worry about score, just check power counter”