two-threads benchmark update

[mgeier]

Instead of using a very large fixed (but kinda arbitrary) buffer size for the benchmarks, I split them in two different scenarios:

• a large buffer that's always large enough to hold all used items, leading to basically no contention between the two threads
• a very small buffer that leads to huge contention between the threads

I also used iter_custom() and Barrier to make the results more reproducible.