Seems anything with s// in a loop doesn't stand a chance.
I played with @- and @+ from s//, but these too, are expensive.
I think this is what you are talking about in the pseudocode?

l_jungleboy: 28 wallclock secs (28.84 usr + 0.00 sys = 28.84 CPU) @ 3550.62/s (n=102400)
leftover: 16 wallclock secs (15.87 usr + 0.00 sys = 15.87 CPU) @ 6452.43/s (n=102400)