If you want to roll the shoulders easily, then a ski bend with the grips rolled outwards will help. So they look like a "Y".
Rolling the shoulders isn't necessarily better. It works for some, hinders others.All you can do is test different positions to see what works for the individual, as eyeball aerodynamics can only get you so far (or make things worse).
If you have access to a PM, then you can do a lot of testing on a trainer (with different stems, risers, pad width,...) to see where power starts drop off or improve, as well as where discomfort/comfort starts to hinder/help. This will give you a good idea of what not to bother with. The same PM can then be used to clock speed at a given power output. These don't need to be done near race speed. 75% is a good area, as it allows good control, as well as not being too taxing as far as testing in the same session goes. Record stack/reach/width measurements for each position and rank them according to which is most aero(greater speed vs lesser wattage). Then during a different session, start with the most aero position and do max efforts in it to see if the position is sustainable, if not, move to the next position,...
If actual track time is limited, and pursuit is being prioritized, them doing stuff like this is quite easy, as it won't take much track time to figure out what works (ideally only two sessions at about an hour apiece). If the stack/reach/width positions are recorded and kept, then tweaking can easily be done as one gets more comfortable with steering in the bars at speed on a track.