0.9 = 0 (not 0.9)
0.9+0.9 = 1 (not 1.8)
0.9+0.9+0.9 = 2 (not 2.7)

That's exactly correct. BOINC 'overcommits' the CPU by 'not more than one complete core'.

Actually it's quite OK behavior.
The alternative is to leave core almost idle if any GPU task in progress.
If GPU app really need full core for support it just provides 1.

---

If GPU app really need full core for support it just provides 1.

But if you, dear reader, decide that the GPU app you've been sent needs a full CPU core in support - you will have to configure that yourself. Whatever the science, whatever the programming language, whatever the observed behaviour - it's still down to you.

BOINC automation won't set it for you.
Project administrators won't set it for you.
Project programmers won't change their programming language for you.

0.9 = 0 (not 0.9)
0.9+0.9 = 1 (not 1.8)
0.9+0.9+0.9 = 2 (not 2.7)

That's exactly correct. BOINC 'overcommits' the CPU by 'not more than one complete core'.

Actually it's quite OK behavior.
The alternative is to leave core almost idle if any GPU task in progress.
If GPU app really need full core for support it just provides 1.

leaving a core idle is more desirable since you often get better production that way without the system tripping all over itself trying to run >100% on the CPU. there's always some background tasks that need CPU cycles, and BOINC doesn't know how to throttle CPU use at resolutions lower than thread count. you can't get BOINC to run 90% CPU on a 4-thread system for example. it'll only do 75% or 100%.

---

Set CPU usage to 1 as FGRP does and you get full core reserved - no probs.
But having ability not to reserve full core by specifying less than 1 is useful.
The choice always better than no choice (and tendency to program in a manner " I know better" usually result in less useful apps, compare BOINC for Android and NativeBOINC when it was supported on OS).

---