round procedure

Hi Manuel & team,

In r5rs spec is described how (round) procedure should behave.
Saying :     ".... rounding to even when x is halfway between two integers"

In older versions of bigloo (3.x) , round was written in scheme and  that is what it does:
1:=>  (round 2.5)
2.0

But in latest versions of bigloo (I suppose 4.x): there is a new C implementation, that does not round like that:
1:=>  (round 2.5)
3.0

I think we should expect an even number (= 2.0)  when 2.5 is inputted into round.
What do you think about that?
Thanks, regards,
Michel