Use a larger input dataset

With this, we should be able to scale the benchmark to 5 timeframes (with 16 events) each.

Therefore with 4 threads, we can have events_per_thread = 4 and up to 5 timeframes.