Do you have periodic changes in the load you are executing that could be writing more? Or... are you running in a virtualized environment like AWS and writing your snapshots to EBS? If so, that might be it right there - network IO issues, noisy neighbor, etc.
You might have better luck using AOF. You would get better persistence 99% of the time, with tunable parameters for how often you would rewrite the AOF.
Oh, can you provide the output of "INFO"? And the first 10 or so lines of top? I'm just looking for the memory and CPU information, not a process list.
- Josiah