Historical node uses single thread

If I type some queries, historical process uses multiple cpu cores but after seconds it uses single core for long time. (almost 10 seconds). Why does this happend?

I configured druid.processing.numThreads and druid.query.groupBy.numParallelCombineThreads but there are no effect.

Please help me.

Could you show thread-dump in that situation? (jstack <pid>)