I've been looking into 'microstuttering' for a while now, since weighing up my last upgrade. My opinion based on what information is available on the net so far, is that it's FUD. OP has linked to forums which link or paste info from the same german article that absolutely every other mention of microstuttering on the net seems to reference. An article on pcgameshardware.de, from February.
I've searched long and hard for some better analysis of this phenomenon, and every mention of it, pretty much all of which are from forum posts, either link back to this same article, or link to a forum which in turn links back to the article. There's only 2 sets of test results that exist backing up this idea, one has already been pasted in this thread, the graph compiled from results on another german hardware site. The other is very artificial looking perfectly straight line versus uniformly wriggly line graph in the article on the first german hardware site.
It strikes me as quite odd that since this is such a simple problem to understand and apparently very easy to demonstrate with a copy of fraps, why every forum thread discussing this is just linking the same results over and over. I've looked through the threads, and the only pasted results from a user in the thread did not backup the original microstuttering claims at all. The hardforum.com post links a graph that a user 'found' - found on the previously mentioned german article. Where's the Anandtech or similar respected hardware site investigation into this, if this is a genuine issue on multigpu systems?
Have another look at those call of juarez figures on the previous page. The numbers show at what time in ms the GPU(s) rendered that particular frame. Note that the multigpu configuration in this graph is actually putting out a lower framerate than the single GPU setup... It takes 1.2 seconds to reach 30 frames, versus 1 second for the single GPU. I can tell you from my best attempt at german that these figures are allegedly 1 8800 Ultra vs 3 8800 Ultra's, the fastest cards available at the time of the article. If that doesn't start ringing alarm bells as to the validity of the benchmark, take a look at the variance of the single GPU results. It's too perfect, they look nothing like the results pasted by users on the other forums, and it's putting out almost bang on 30 fps. I'd take a wild stab in the dark and say it's running at a framerate cap.
The original article describes microstuttering as an inherant problem with all AFR multigpu solutions, and though they only have 1 questionable graph of 1 card in 1 game, they are happy to say that this is a problem with all hardware in all games and has no solution. It's safe to say just from the responses in this thread, that this is a load of FUD.
As far as i can tell, the only reason this article is still being propagated is because it's far too easy to associate the word 'stuttering' with absolutely any type of game related performance problem you may have. So anyone having any type of performance problem on multigpu who reads the article suddenly has their answer. Low on video memory and getting little pauses as textures are loaded from ram? Microstuttering! Low on ram and paging from disk? Microstuttering! Framerate under 60 in crysis? Woa hang on you're using SLI, that must be microstuttering. This is why you get things like this:
http://www.youtube.com/watch?v=-5SXeR0torc pasted as an example of the phenomenon, when the slightest bit of thought would tell you that short noticable pauses every couple of seconds visible in a youtube video cannot possibly be the microstutter issue the german articles are talking about.
That said, even after seeing it in action running very nicely, i'm not going for a crossfire setup. Nothing to do with 'microstuttering' i just like my double monitors, multiple GPUs have many issues, but i'm not at all convinced that this is one of them.
