Some observations:
1. MIPS doesn't tell anything of the real performance. "MHz" of the emulated CPU tells even less.
2. Synthetic benchmarks are useless under emulation.
3. JIT will optimize away most "timing loops".
For any sensible real world results:
A) Use real life tasks, such as compression or decompression of large dataset.
B) Time the operation from outside of the emulation.
C) Run the benchmark multiple times. Take average of 3 best results.
