The best way is with the horizontal, as was said, if its rigid ie very heavy.
It will be very fast, and very accurate, one setup/go.
Passes will depend, maybe 2-3 or more at 3 hp, depending.
The 3hp indicates its not very heavy(maybe ?), but I would still plan for horizontal only.
A proper rougher is very efficient.
Use a modern one.
They can remove 3-5x more material / hp / time than old school stuff.
Which one ?
Depends.
A corncob type rougher in solid carbide is by far the best, if the cut /machine is suitable.
Ie given enough rigidity so it does not chatter, the corncob solid carbide will be fastest MRR, with best results, in least time.
It will be expensive.
You said 50% shaft, ie 3" width of cut.
Expensive. Might cost 500$. Or more.
A cobalt or HSS rougher may be 1/2 to 1/4 the cost.
So a small rougher might be best overall, and very much cheaper to replace.
Otoh..
A smaller endmill rougher, about 10-12 mm, would be the best option, imho, on modern high speed machines.
It would run at 10-20x faster feeds, and do many passes, much faster than a single large, heavy, expensive, rougher.
Maybe 300m-6000 rpm. TLAR method, just from people using them on more rigid stuff.
At 3000 rpm, on 300SFM, 0.5", 0.002 /teeth, 3 flute.
= 3000*.002 *3 = 18 "/min.
If the flat is 1" in length, it takes 20 secs to do, and 6 passes / depth at 0.5" endmill rougher D.
The *right* answer depends on machine, rougher, how many pieces, if $, time or production qty is more important.
The best answer with all data will still vary by more than 100%.
Ie with the best possible solution, someone can make it in half the time/cost.
Edit.
Possibly the best option.
Slit it with slitting saw. One pass. Cheap.
Then clean it with minimal cut using anything, a facemill or even a flycutter.