I spent some time investigating this issue and I think the problem is hardware failure due production problems in the heating dissipation system. Not design problem but pure incompetance in manual assembly of the components PLUS a unstable GPU.
I found some concerning pictures of the MBP 15 early '11 tear-down, where we can view things that are away beyond the lack of quality control, such as thermal paste spread like "butter over the bread" all over the processor's board, as well as losen ZIF connectors, etc...
In terms of design, GPU and CPU shares the same cooling pipe and the graphic switching processor has it's own, which is different from previous mid '10 MBP logicboard design.
So, I think the GPU is overheating way faster than CPU and, when fan throtles in (controled by CPU temp) it's not enoght to keep GPU temperature threshold and cause GPU to fail.
After GPU fails, it corrupts some part of the NVRAM which prevents the machine to boot up even after temperature is OK. The way I found to boot up again is by running Apple Hardware test. Besides the test itself doesn't points out any hardware failure (even running the test with display messed up!), I found that after running the test I was able to boot the computer again. But it don't works all the times and I need to run it twice.
Just after the boot, installed gfxCardStatus to prevent the machine to change from Intel (integrated GPU) to AMD (discrete GPU) and installed SMCfanControl to throtle the fans up to 5500RPM.
After, I have (so far) a stable machine, even after changing from integrated to discrete GPU (by disabling gfxCardStatus and opening Photobooth and VMWare running WinCAD), and after many boot cycles, but with downside of higher fan noise.
Of course IT IS NOT a solution, just try to make a point that problem is not related to FW or SW problem, neither from a single machine but affects all machines with this HW and the time frame for this problem to occurr is related to some very subjective issue: basically the amount and the way of thermal paste was applied to the processors and heat sink at each individual machine and the tolerance to temperature of GPU.
This of course may not apply to all machines.
_____UPDATE
Now I was looking at iStat widget and GPU temp is 43 deg and GPU Diode temp droped from 25 to 3 to -128 (whow, ...). Even not been a reliable tool, istat takes information from the on board sensors so... something is not right..
Follow some pictures

