Yes the paper compares it to Core2Duo. Which is about 107mm2 for 45nm. 16 cores and add the fixed function units and you'll get 200+mm2. That's probably why they'll go straight to 32nm. Larrabee need about 48-64 cores @ 2 GHz to be competitive with AMD and NVDIA future offering. I think if Intel can't secure a console contract, the project will be canned.
I read the pdf again, you're right Intel hypothesis was that they would be able to fit ten cores within the same die space as a Core 2 duo with 4MB of L2 cache, they also stated that ship would have about the same power consumption (between they don't say per clock which is unlikely).
I found the same references as you here:
http://www.pcgameshardware.com/aid,665558/Intel-Core-i7-Nehalem-CPUs-reviewed/Reviews/
Core i7_45 nm_263 mm²_731 M
Core 2_45 nm_2x 107 mm²_2x 410 M
Core 2_65 nm_2x 143 mm²_2x 291 M
The aforementioned core2 include more than 4MB of shared L2 cache 6 indeed.
It's not clear if Intel included the fixed function in their calculation but it's not that relevant either the presentation I watched by yesterday night was from the siggraph 2008 and they used the same figures in even earlier presentations. Basically I think that it's a gross approximation and that we should not read too much in it... Anyway I'll do
Using Intel gross estimation a 20 cores version (8MB of L2 cache) could take from slightly less than 200mm² to ~270mm².
Best case: 2xCore 2 minus 2MB of cache(should be under 100mm²)
Worst case: nehalem => 4cores+9MB of L2/3 cache+extra logic MC mostly ~270mm²
the 16 cores renditions would land anywhere between 160mm² and 216mm².
But I think that 216mm² would be a failure for Intel or in density or in transistors count per core (~40millions transistors) and the project could be canned.
But I would read too much in Intel waiting for 32nm, Intel has pushed back a lot of projects till 32nm process availability, I think it has more to do with die size and space thus costs.
I makes sense for Intel to avoid pushing a power hungry costly and low yields >300mm² monster.
Intel can wait some more both in the GPU and in the GPGPU market especially as extra time put on the software side on thing will have a huge impact on how the product will be percieved.
I don't think that the project will be canned if Intel doesn't make larrabee the heart of one of the next generation console, it's could still rule the show in the GPGPU realm even if too big/not competitive with Nvidia/ATI on the graphic side. But then it's more an Intel vs Intel decision ie larrabee might not be allow to stealth some market shares from others potentially more profitable Intel divisions.