Why does my Xeon E5450 engage thermal throttle so easily?

It runs at full speed for 2 minutes then when it hits 35 degrees under full load with Prime95 it changes the multiplier from 9 to 6.

That's not a high temperature at all, but interesting to note that it's a very cold winter day, and on a hot summer day it would downregulate at 50 degrees which is at least close to the 67 degree max rating. But it still seems really conservative, and the difference in reported temperatures suggests something wierd is going on.

I have a huge heatsink (skythe infinity) with a small blob of artic silver. Only thing is I'm using slow fans (push pull) for ultra quiet operation. When I was playing around with other heatsinks and fans, I never found a way to keep this chip cool enough to run at max speed full load continously for more than a couple of minutes.

Just looked up the TDP and it's 80W, for some reason when I built the system I believed it was 120W, so I went to town on creating air ducts inside the case with aluminium duct tape so the air travels on a defined path from the front of the case (where there is an additional intake fan) to the back where warm air is exhausted. So the air moves slowly but efficiently, being drawn into the vent on the front (which I can feel happening) 50cm from the warm side and at 180 degrees. The case is from an old SGI Intergraph Workstation, not some junk box case, it's multiple kilos of solid steel.

Motherboard socket is rated to 136W. PSU is a Corsair RM750.

Also I am overclocking but observed exactly the same behaviour at stock speed, thus concluding that I might as well overclock if I can only get 2 minutes of high speed it might as well be as high as possible.

So is my cooling inadequate or is this just what I get for buying one of those stickers to make the Xeon work in a cheap desktop mothermoard?

Maybe the VRMs are overheating if air is stagnant around those because of the air tunnel you created. Try to remove the tunnel and blow some air on those parts with a fan of even yourself with a piece of paper and see what's happening. I think you might solve the problem this way. Also if you can try to undervolt the CPU even at stock speeds and see if combined with VRMs cooling the speed stays stable.