Performance regression with new scheduler

[dignifiedquire]

It turns out that #733 introduced quite the performance regression on various http benchmarks, > 50% on https://github.com/stjepang/minihttp/

I have tracked the issue down to this commit: f960776

If I revert it, all tests still pass and performance improves to be even slightly faster than 1.5.

@k-nasa could you shed some light what exactly you were trying to achieve in this commit and why you introduced it?

[dignifiedquire]

@k-nasa I figured out you were trying to match the original PR, I think you didn't match the exact logic, which resulted in the perf regression. Please checkout out #747