I did some quick tests on Cinebench 15.0. and at least the CPU throttling seems to be a pretty straight forward thermal issue.
10 CB iterations per 1 test. Undervolt was managed by ThrottleStop 8.60, Speed Shift had been enabled on all tests (was off on default settings), with a value of 128 so some mild underclocking compared to stock could be expected on all cases.
The laptop stand had 2 60mm fans which I had based on the upper part of the laptop, just under the CPU/GPU.
Test 1 - No (laptop stand) fans, no undervolt
CB scores:
755 711 709 688 690 641 623 628 638
mean : 676
Test 2 - Fans on, no undervolt
742 673 690 696 678 555 700 692 694 695
mean = 682
mean ignoring outlier (555) = 696
There was 45 min of pause between Test 1 and Test 2 (went to the store) so the system was cooled down. Still it seemed to start to throttle faster than Test 1. No idea why. The floor seemed to be higher though.
Test 3 - No fans, -60mV undervolt
733 745 732 660 705 714 738 730 701
mean : 718
Test 4 - Fans, -60mV undervolt
751 747 744 753 754 752 750 683
mean : 742
The 10 iterations is probably too little to see the system stabilise fully, so to see the real gains much longer tests should be done. Anyways, it seems like that if we trust the Cinebench scores it is possible to gain ~10% of sustained multithread performance on a 10min intensive workload by using a laptop stand with fans and applying a relatively conservative underclock. A ~6% performance gain was seen with just the underclock. Keeping this in mind I atleast am going to repaste the CPU when the warranty runs out, and probably going to push the underclock a bit more.