A possible problem is neither of the three. Usually, GPU (as well as CPU) is connected to the heatpipe through a thermoconducting block (copper or aluminum, sometimes a weird blue thermal rubber), with a tiny quantity of thermal grease to fill in microscopic cavities which could lower conductivity. Too little grease leaves some spaces open, and too much grease acts as an insulator; both can easily reduce thermal conductivity enough to raise CPU/GPU temperatures by 10-20 °C.
Sometimes, the air flow also is modified by foam blocks that can easily get dislodged during servicing (the "WTF is this stuff for?" effect). Then, less air is collected from the hottest areas, replaced by not-sufficiently-warmed-up air from nearer inlets, leading to temperature increases.
I can say that after replacing my old and cracked thermal pad with a copper shim and two tiny drops of liquid metal (dangerous as all hell to work with in a computer (1)), my GPU went from around its usual 85 °C to 68-70 °C when watching videos. The heatpipes were OK, but not enough heat was getting through to them from the GPU.
(1) "liquid metal" is actually an alloy of gallium, indium and tin. It is one of the best thermal conductors that can be used in this situation because it's a metal and, hey, it's liquid. BUT being a metal, it can short circuit a motherboard with absurd ease. Also, gallium (as well as the much more toxic mercury) can poison some metals such as aluminum: never place it in contact with aluminum, and clean immediately any drops that might have come in contact with an aluminum chassis. A telltale of gallium contamination is grey-black smudges that apparently come from nowhere (they actually come from tiny gallium droplets).
