TMU's follow the shaders...so it will have a full complement of them to go with its 240 shaders if that is indeed the configuration.
The GT300 story makes no sense. The 55nm GT200 is still a big chip at 470mm^2. Their next monster has to be on 40nm out of necessity. And where are all the 40nm GT2xx derivatives to replace the G9x series cards?
Remember Arun's theory on GT300's memory bus width and RAM type, etc ?
That would leave a lot of empty space in the die, ready to be filled with... something.
Besides, making huge chips isn't exactly new at Nvidia (NV30, NV40, G80, GT200, etc).