I have a Sapphire 9100 card very similar to the one discussed in http://forums.silentpcreview.com/viewtopic.php?t=5533
. The fan is indeed amazingly noisy once you've quieted the CPU and PSU. Other threads contain opinions about running fanless or with a different HS (mine works ... mine doesn't .. this HS is good ... etc), but little in the way of facts. So I decided to measure some temperatures, using the nifty Enermax UC-A3FATR2 (see http://www.enermax.com.tw/uc-a3fatr2.htm
for details - a useful and inexpensive experimental gadget). The thermistor with some thermal compound on was slipped under the fan wires, which is about as close as you can get to the chip without demounting the fan. The other sensor was in free air inside the case, a little way below the PSU. I have a free slot next to the 9100 card. I wasn't running any games etc while this was going on.
Room ambient : 26C
Case internal : 31C (MB and Enermax)
GPU (fan unplugged) : 53C, delta=22C
GPU (fan operating) : 36C, delta=5C
[added later after installing Panaflow ...
GPU (fan unplugged, Panaflo 80mm 24V-M1A @12V) : 45C, delta= 13C
GPU (fan removed, Panaflo 80mm 24V-M1A @12V) : 45C*, delta= 13C
* slightly different sensor position, since the fan was holding the sensor on]
This isn't an ideal experimental setup, but a tentative conclusion seems possible. The operating range of the card is 10C-50C : assuming the same delta the maximum safe measured temp would be 55C - just above my 53C. So running this card fanless, with no HS mods, looks just about OK provided you have a case temp in the low 30s. The setup I used is probably worst-case, since the unplugged fan is still attached to the HS and the plastic frame must have a significant thermal resistance as well as impeding what airflow there is.
However, it does seem a bit marginal, especially on a hot day. The HS is pointing down in a standard ATX case, so I doubt if additional pure convective cooling will be very effective - plus the aggrevation and cost of replacing the HS. But given the clear effectiveness of the tiny standard fan, I expect any reasonable amount of airflow will do the job. So I think my solution is going to be to remove the existing fan, leave the HS in place, and put a fan running @5V over the top on a bracket.