http://www.zorro-trader.com/manual/en/numsamplecycles.htm
"When NumSampleCycles is set to a number n > 1, the simulation is repeated n times, and every time the bars are resampled with different BarOffset values."

Intuitively I would expect training runtime to scale linearly with NumSampleCycles, but this is not the case. A training for a strategy NumSampleCycles = 3 takes more than 3x the time as the same strategy with NumSampleCycles = 1.

Can it be explained?

Example
Add set(TESTNOW) and NumSampleCycles = 1 or 3 to Workshop_6_2
Then time Zorro -train Workshop_6_2
NumSampleCycles = 1: real 2m31.256s
NumSampleCycles = 3: real 8m2.807s

The non-linear effect is more pronounced in other situations.

My guess is that 3 sample cycles need to allocate three times the memory, which leaves less free memory. This can cause higher memory fragmentation and more swapping, and thus generally a slightly slower execution.