Michael,

> So all in all (based on 1 day of observation ;-) ) I think the sweep
> did take a lot of res. and memory.

Seems that you found the problem.
Good.

> Another thing I tried (on a diff. DB, a smaller customer) was forcing
> a disconnect from the DB and a connect to the DB every once in a while.
> This customer now has a gab between 50 and 400. Usually it was
> somewhere between 500 and 6000.
>
> I think I'll try that on this big customer DB as well.
>
> Anywas with any comments on this ?

Yes.
Try to find out which transaction(s) is (are) running that long without
being committed and try to change this.
Don't use SavePoints or CommitRetaining if there's a chance to use commit.

I would prefer this one over disconnecting and reconnecting to the db ...

Luc.