Two things affect the degree (aka frequency and severity) of cpu-bottlenecking:
1) The level of cpu-dependency of the test (game/bench) you're running.
Loosely defined, this means "How much work does the GPU require of the CPU per frame rendered?". Various applications vary WILDLY in this regard. Not only is every application different, but it's an ever-shifting variable over the course of every test you ever run, and
2) The graphical difficulty level of the test (what game/res/settings) relative to the power of the graphics solution.
This is because the graphical difficulty (obviously) impacts the FPS at which the test is running, and hence how OFTEN the GPU makes requests for data from the CPU (see #1).
Since FPS is always going to vary, and the cpu-dependency is also always going to vary, depending on what's going on onscreen, the phenomenon of cpu-bottlenecking is a very dynamic situation.
Basically, on *every* frame *ever* rendered on your system, either the CPU or the GPU will be the bottleneck.
The bottom-line all this is, the faster your CPU is relative to your GPU(s), and the higher the level of graphical difficulty (and hence the lower the FPS), the less frequently it will be the case that the CPU will become the limiting factor to FPS, and the less severe the limitation will be when it occurs.
And there is no 'magic number' afa a clock speed goes ^^^ see my previous paragraph.
To answer the specific scenario at hand ... I think any present AMD setup is a bit underpowered for SLI580's, but you can of course still create gaming/testing scenarios where the BN would be mostly invisible ...