How long do you plan the chip?
Do you overclock?
How much data safety do you require if you overclock?
%%
Not overclocked, the FX-57 will be faster for non-multithreaded applications for now, but not by much. Most benchmarks don't show the extra Megahertz to be very noticable.
And not only does the FX-57 start out only slightly faster, it might lose than speed advantage during the time you use it. If you use the chip you buy now for 2 years, then most applications will be out with multithreaded versions by then, giving the 4800+ a clear edge.
However, if you overclock, the FX-57 is better. First of all, having only one core it has a better statistical chance to reach a given overclock, since in the 4800+ the core which overclocks worse limits the whole thing.
Then, the FX-57 has the unlocked multiplier which enables you do overclock and test safer and in a more straightforward manner.
If you are willing to overclock you likely also use faster than DDR400 which normally doesn't do much for AMD64s, but in this case might help the FX-57 to show off its extra 200 or 400 Megahertz.
%%
Last but not least if you overclock the FX-57 is easier to test, to verify your overclock is safe. While running two Prime95 or SuperPi will do some basic testing the exact interactions of two overlocked cores are not very well understood. Alternate overlock testing such as FreeBSD's `make world` which I use to test hardware are not easily multithreadable and won't reliably load two cores. Running two at the same time chances the whole system dynamics and might invalidate the test run.
I also noticed that the final point of making an overclocked single-core chip in my testing sometimes is a `make world` simultaneously with a Mozilla build. Now transpose that to the SMP situation, how do you reliably put this dual load (actually alternating load per core) reliably on each core, at the same time?
Overall the FX-57 will not only overclock higher, it will do so much more safely.