The latest 3.08 CVS solves the problem (for GlSurf at least).
When I have the too Glsurf thread (on computing, one for drawing GL (I 
keep pressing a key to rotate the surface) and I launched povray from 
Glsurf for a rendering, then I get one third of cpu each which is perfect)

Just a last question:

Now I saw that for "non broken" sched_yield, the call is still present.

Are you sure that releasing the mutex is not enough to tell the 
scheduler it may be a good time to give some cpu to another caml thread 
blocked on the same mutex ?

But I am sure you tested that too and this is why the call is still 
there when possible ;-)

-- 
Christophe Raffalli
Université de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tél: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
---------------------------------------------
IMPORTANT: this mail is signed using PGP/MIME
At least Enigmail/Mozilla, mutt or evolution
can check this signature
---------------------------------------------