A simple example: RosettaCode's Universal Turing Machine (5-state, 2-symbol probable Busy Beaver machine from Wikipedia) In V9 time = 225 seconds, in V10 322 seconds (Windows i5 desktop).

Not much to do but hope for better performance. I think I hear more disk swapping but that's not measured (yet).

I gave up with the Benchmarks some time ago. The tests are not run for the new release so you're always comparing apples to oranges.

Indeed. Some of that is due to MKL getting in update (linear algebra, FFT, etc.) Do we know if the benchmarks included with v9 and v10 are identical though?

My general impression has been that most things are faster in v10, but some examples (like this one) show that some other things got slower. It would be nice to be able to track down what exactly is causing the slowdown.

I can also reproduce a longer processing time (about 20%) with my Windows 8.1 machine (4 cores/8 logical processors): 186s (V9) - 228s (v10)

The web page has the exact code I uploaded. Put Timing[ ] around any driver/test. The last one is the long-running example.