It’s currently the maximum time of all the threads. As mentioned in Make the solveequihash benchmark more representative of solution latency / add another benchmark for throughput · Issue #1191 · zcash/zcash · GitHub , this is not what we want since it isn’t a good measure of performance. Until this is fixed, I’d recommend just ignoring the multi threaded benchmark option, and instead using genproclimit and measuring from logs the number of solutions that are difficulty-tested in a given time. (This is essentially what the throughput-oriented benchmark will do.)
Here’s how to get the log output: Zcash daemon logging settings? - #3 by str4d