As I just wrote in a thread in CPU subforum...
The peak GFLOPS possible of a chip is only down to three things: the clock speed, the number of cores, and the instruction sets.
Example:
Your 2500K has 4 cores, is clocked at 3.3 GHz (say), and possesses AVX and you're using v2.52 on Win 7. The peak possible GFLOPS is then 4*3.3*8 = 105.6. (That 8 becomes a 4 without AVX.)
The more memory you use in IBT the closer you'll get to this value, but the longer each loop will take. Most likely you'll see it max out around 80%.
In summary, to stress test at stock (not sure why you would do this) you must use sufficient memory to get about 0.8*105.6 = 84.5 in the GFLOPS column.
So to help you with your question - I suspect that the GFLOPS reported by IBT is far more dependent on the memory usage than "degradation" in the chip, if it exists at all. "Degradation" is more likely to effect stable overclocks than computing power at stock, IMO, if it exists at all. 93 is perfectly fine.
.
