How do you know it uses 32 threads? I mean which tool do you use to determine that?

BTW, except mining treads, cpuminer runs several other auxiliary threads (api tthread, stratum thread, few other threads depending on specific version).
So if you see 32 threads shown overall, the mining threads could be even less.

Also, more threads does not equal more productivity, because this algo uses memory bandwidth a lot, so that could be the bottleneck, not the cpu power. You may want to experiment with "-t 18" to see if this gives you anything different from "-t 36" in terms of hashrate.