I've got this same problem (P105-SP6022 w/7900GS GO) and, over the last week, have tried various fixes to get around the acpi=off trick (because a laptop with no ACPI isn't that fun). Here's what I've found so far:
1) The fan can't be controlled by nvclock (at the moment) because the card uses a Maxim MAX6659 sensor. That sensor only provides support for temperature readings and not for fan control (http://www.maxim-ic.com/quick_view2.cfm/qv_pk/2578)
2) Whatever "thing" being used to control the fan is not detectable by lm-sensors and no amount of googling has turned up anything useful on this front.
3) I disassembled (and fixed) the DSDT and noticed that it does actually recognize that it's running on linux however it takes no action based on that. The only optional actions it takes are based on Vista (which it calls "Windows 2006") and booting with acpi_os_name="Windows 2006" has no effect.
4) There is something that, in my ACPI novice eyes, looks to be a VFAN device (which I read as video fan) in the embedded controller section of the DSDT but, I have no idea if that's actually the case and I'm afraid to even try to figure out how to read from it.
That's all the info I've found so far. I'm getting into Super Speculation Territory here as I know approximately nothing about driver development but, my guess is that on Windows, the modified Toshiba nvidia driver is constantly polling the MAX6659 sensor and then writing something into it's own embedded controller (maybe at the address of VFAN) that is controlling the fan.
So, I don't *think* it's a problem with the nvidia drivers but instead with Toshiba simply not telling nvidia how to control the fans they are using on their nvidia based laptops. I'd petition them to try to get the information but, I don't think I'd have much luck.
Has anyone at nvidia talked to Toshiba about why their GO series cards don't work on linux? Are they simply not willing to share the address at which one has to write 0x00 and 0xff to turn the fan on and off?