Performance regression in 1.8.0 vs 1.7.3

[jzakiya]

In running my standard apps to benchmark against 1.8.0 I found a regression when running this code.

https://gist.github.com/jzakiya/2b65b609f091dcbb6f792f16c63a8ac4

This is one of my parallel twinprime sieves.
I think the regression is in the concurrency changes from 1.7.3.
Notice not only do the times get progressively slower for larger inputs,
the stripped binary size for 1.8.0 is over twice the size for 1.7.3.

For other code that doesn't use concurrency I don't see this speed|binsize disparity.

This is my system.

➜ inxi
CPU: 8-core AMD Ryzen 9 5900HX with Radeon Graphics (-MT MCP-)
speed/min/max: 1249/1200/4679 MHz Kernel: 6.1.24-pclos1 x86_64 Up: 2d 16h 43m
Mem: 14734.9/39489.3 MiB (37.3%) Storage: 953.87 GiB (12.6% used) Procs: 413
Shell: Zsh inxi: 3.3.26

    Inputs   |  1.7.3 |  1.8.0 |
-------------|--------|--------|
    1*10^11  |   1.23 |   1.46 |
-------------|--------|--------|
    5*10^11  |   6.18 |   7.49 |
-------------|--------|--------|
    1*10^12  |  12.93 |  15.67 |
-------------|--------|--------|
    5*10^12  |  74.05 |  86.49 |

    
             |    1.7.3 |   1.8.0   |
-------------|----------|-----------|
 Binary Size |  487,872 | 1,000,776 |