With the STI strut brace being introduced, I thought it would be good to bring them up in the discussion... since it's theory and all. Much of it is copy-pasta from that thread, so if you've read it there, disregard.
I'll use this discussion as a source:
http://www.e30m3project.com/e30m3per...bar_theory.htm
Strut bars have 2 purposes:
1) Stiffen the tower to tower lateral movement (tower to tower bar)
2) Couple strut flex to a neutral chassis position (tower to firewall bars)
Many people discuss the chassis surrounding your suspension as being a parallelogram. While it's not completely true, it provides a useful illustration.
Point 1 makes the top line in the parallelogram more rigid. In most cars, the other 3 lines are already very stiff. This turns your trapezoid into a parallelogram (a good thing).
Point 2 makes the angles at each corner of the parallelogram more rigid.
During compression, the strut towers want to bend inward. A strut bar is put into compression and resists this quite well.
During cornering, one strut tower wants to move inward, while the other moves outward. A strut bar ties these deflections together, stiffening the one that deflects more (like a sway bar does for vertical motion). Since the inside tire's camber is less relevant to maximum grip, you sacrifice some deflection to improve the situation on the outside tire.
All this is dependent on the struts flexing. Modern chassis design is rather rigid, so it's not as important as in an 80s Civic or my RAV4 (an SUV version of the Corolla).
When I installed my strut brace (1.5" square tubing w 1/4" plate on each tower), the handling limits did not really improve... but the consistency of when the tire broke loose was much better. Strut flex begets more camber and grip, and in many cases is not a bad thing. I value the stability and consistency more than ultimate grip. Go kart vs. muscle car, if you will.
Point 2 is irrelevant if you consider the front and rear suspension as independent parallelograms. They aren't... so preventing either from differing too much from the unstressed geometry is important. Also, triangles are stronger than parallelograms, so tying into the firewall (however flimsy it may be) is still good for structural rigidity of all components in the party.
Modern cars indeed make strut braces pretty irrelevant, since the structures are inherently more rigid. That said, if the M3 guys estimate 0.5 degrees camber change due to tower flex, I'm willing to add some reinforcement to prevent it. Some brave soul can install a strain gage on their cusco bar and see that there are indeed forces through it... but the magnitude of the deflection just isn't what it was 30 years ago. Bushings are another story for another day though.