I'm back with some preliminary testing.

CPU: i7-960
GPUs: 3x GTX260
 
I'm currently pretty busy, but I got the system to post and boot. I took some rough seat-of-the-pants measurements in Metro 2033 Redux.
Currently running high settings with PhysX turned on. Walked the same stretch of game play for each run. Some combat. Eyeballed FPS with Steam's FPS overlay. 1600x900.
 
All four components water cooled in serial w/2x 360mm radiators. Only 3 fans installed on the top radiator currently. Temps maxed out at 60C for CPU and 50C for all GPUs.
 
Single GPU: 40-60FPS

• Pretty stable, no stuttering. What you would expect from a single card.

2-way SLI (x16 and x8, x4 dedicated PhysX): 80-110FPS

• Some hiccups occasionally, but overall, pretty smooth. Pretty close to a single high performance card experience.
• Saw ~95% utilization on cards.

2-way SLI (x16 and x4, x8 dedicated PhysX): 80-110FPS

• Not much to say, felt similar to the last setup.

3-way SLI (CPU set to PhysX): 70-100FPS*

• More noticeable stuttering, but felt pretty similar to the 2-way SLI runs.

 
I still haven't tried installing the card in the other x8. I need to get a longer male-to-male fitting for the lower GPU and the bridge with wider spacing for the bottom two cards. I'll come back when I have some more time and the parts on hand.
 
Some more controlled testing across a few games would give a more accurate picture, but I think the 2-way SLI results say enough for my purposes.

a_scerba
I'm back with some preliminary testing.

CPU: i7-960
GPUs: 3x GTX260
 
I'm currently pretty busy, but I got the system to post and boot. I took some rough seat-of-the-pants measurements in Metro 2033 Redux.
Currently running high settings with PhysX turned on. Walked the same stretch of game play for each run. Some combat. Eyeballed FPS with Steam's FPS overlay. 1600x900.
 
All four components water cooled in serial w/2x 360mm radiators. Only 3 fans installed on the top radiator currently. Temps maxed out at 60C for CPU and 50C for all GPUs.
 
Single GPU: 40-60FPS

• Pretty stable, no stuttering. What you would expect from a single card.

2-way SLI (x16 and x8, x4 dedicated PhysX): 80-110FPS

• Some hiccups occasionally, but overall, pretty smooth. Pretty close to a single high performance card experience.
• Saw ~95% utilization on cards.

2-way SLI (x16 and x4, x8 dedicated PhysX): 80-110FPS

• Not much to say, felt similar to the last setup.

3-way SLI (CPU set to PhysX): 70-100FPS*

• More noticeable stuttering, but felt pretty similar to the 2-way SLI runs.

 
I still haven't tried installing the card in the other x8. I need to get a longer male-to-male fitting for the lower GPU and the bridge with wider spacing for the bottom two cards. I'll come back when I have some more time and the parts on hand.
 
Some more controlled testing across a few games would give a more accurate picture, but I think the 2-way SLI results say enough for my purposes.

Have you tested with 3-way sli but leave Physx on auto?  Physx on the CPU will slow things down and cause stuttering as you've found out already.

bdary
Have you tested with 3-way sli but leave Physx on auto?  Physx on the CPU will slow things down and cause stuttering as you've found out already.

No... I'll give that a shot. I understand SLI is pretty touchy. I did try setting it to the third card, and performance was pretty abysmal. 10-20fps.

Tried setting PhysX on auto. It selected The 8x card. Performance was bad. 10-20 fps. Bad stuttering. Tried setting it to the main x16 card. Similar result, but this time the game crashed about a minute in.

I guess you just have to try every possible combination and then run with whichever gives the best results.

bdary
I guess you just have to try every possible combination and then run with whichever gives the best results.

Yep, seems 2-way with one dedicated to PhysX is the best currently.

Gonna try some other games soon.