Sort Performance Degradating when nthreads bigger than machine cores

I wonder why sort operation is sensitive on nthreads.
Test on my machine, when nthreads is just one more than the cores.
it will occur Significant performance degradation

29 threads vs 20 threads 【machine is 28 core】


# machine：

cpu：Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz
cores：28
processor : 27
vendor_id : GenuineIntel
cpu family : 6
model : 85
model name : Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz
stepping : 4
microcode : 0x2000043
cpu MHz : 2000.000
cache size : 19712 KB
physical id : 1
siblings : 14
core id : 14
cpu cores : 14

MEM: 263412544k