Adjusted loops_per_ms calibration to avoid infinite loop on some machines

[bartek-poleszak]

When I've tried to run interbench on one particular machine (i7-4810MQ laptop, Ubuntu 16.04 with mainline v4.15 kernel) I've noticed that it is unable to calibrate loops_per_ms, because run_time variable at line 1029 had value about 30% higher than expected.

This pull request resolves (works around?) the issue, but I'm not sure if it does bias benchmark results, so I'm putting it under review.