Can this be true? I over clocked my MSI factory over clocked gpu via Nvidia tuner. I just set up the core and ram speeds. And no issues for now. But it is clocked really high now ( 760 core and 2140 mem). GPU-z says the same.  Why do I think it's strange? trying to over clock core to 750mhz via ATITOOL caused non-positive problems lke system reboot etc.
And card is not hot at all 40 idela 55 load. Is ntune such a good program? Or am i missing something? Can there be that this overclock  is somehow only "on paper" ?

Dunno, grab Rivatuner and monitor your vidcard to see if the clock speeds are close to what they are set at.

i used ntune for awhile - seemed to work great to oc my cpu/ram and video card only when certain applications (games) were launched.  however, once i upgraded my cpu and gpu, the profiles would not save and i had to manually set it each time.  i then started using rivatuner for gpu and just oc'd cpu in bios and that works well.  ntune seemed accurate and worked great for me for awhile.

Because, ATItool is, you know, meant to run on ATI cards, instead of the Nvidia card (assumption) you justed used Nvidia tuner on?

And yeah, Rivatuner is the best. Works on both brands too.

Yeah, riva tuner monitoring shows same clocks. M and does anyone know how far can possibly be MSI 9600gt over clocked? (2 pipelines + fan).

 
Ntune is a buggy program IMO> I would use rivatuner. However if Ntune doesnt crash you might have some luck. I would always keep your shader/core linked for best results. I dont know how high you can go with you particular card. I would start by moving core slider up 20mhz increments and running ATI tool until you are stable 1 minute. Then do the memory at 20mhz increments until you are 1 minute stable with ATI tool. If you are not a fan of ATI tool than you can use 3dmark06. It usually crashes early on if you have artifacting. Or if not you'll be able to see artifacting if you are paying attention.  
 
Set your speeds via ntune or rivatuner than use ATI tool to see if it's stable. Is that what your question was? Your question was kind of worded funky.

you better off with Rivatuner as it overclock Nvidia card very well.a better program to run is RTHDRIBL with high multisampling.that program is much more effective then ATITool's 3D model.as it uses HDR rendering which is more stressful to the GPU and since you can up the resolution and multisampling to use more memory to stress the ram at the same time.
 
i sse many people hit 760 on core and 1850 shader with 1050 ram.but be careful as its a single slot fan so heat is a problem.you can overcome that by running fan at higher speed(min 75%),as long as you can bare with the noise it makes.it you got a dual slot model i have seen it overclock near 800mhz and 2000 shader.
 
ABout nTune is good when the who setup is on a Nvidia platform.as it can manage not only the graphics card but also all the overclocking and other manage feature on the motherboard when use any nForce mobo.

Until you test it in a game you won't really know. Then again, maybe you have. If you are stable with no artifacts in extended 3d then I would have no worries.

Is there any evidence to show that keeping shaders linked versus unlinking and shooting for higher shader clocks has any real benefit? I know, it's only somewhat related.

in any processing unit that higher the speed is better so unlink will offer better performance if one cant overclock as high where as the other parts still have room to overclock.

It depends more on specific code than raw processing power.
 
If for a specific block of code the bottleneck is the shader processing speed it will not gain on speed if you increase just the core clock. The same applies if it's the other way around.  
 
The 2.5x factor is a general optimization by NVIDIA I think. And since most games are optimized for their cards it's quite hard to find general improvements by changing just one clock.