[Julian]

Interesting discussion. A few points to consider before I'll show you all one way (the 'correct' way) to test for torsion. No cheating and scrolling down to the photo now!
What is the point of chassis stiffness? Well, for a car with functioning suspension (not talking about weird stuff like dragsters, F1 or similar), it allows the suspension to do the job it was tuned to do, and not the chassis flexing excessively so as to dilute each minor suspension tweak. Given the nature of cars we are talking about (road/circuit race), it is important to maximise chassis stiffness, within reason. Meaning stiffness vs weight is a compromise, like all things. However, some intelligent design will allow maximum stiffness per given weight.
The other point of chassis stiffness is that the registration authorities here in Australia tend to require documented proof! So it takes on a whole new importance down here!
So what are some realistic numbers for actual kit cars? Well, it just so happens I have a few results to hand. These are real results, not internet hearsay. Having said that, the results list posted earlier in this thread seem realistic to me.
OK - extensively modified (extra bracing) Locost +100 chassis 5400 Nm/deg
Locost type chassis with live rear axle (32 mm sq tube used, not 25 mm sq tube as usual) 4300 + Nm/deg (test was performed without dial gauges at rear (fixed) axle line) so this value is a bit conservative. Could be 5000 - 6000 Nm/deg in reality
Similar chassis (same 32 mm tubing) but IRS design 5950. Same conservative testing so probably 7000 - 8000 Nm /deg in reality.
Cobra replica chassis (kit car manufacturer supplied) 8000 Nm/deg
Cobra replica chassis (scratch built) 6500 Nm/deg

These are actual tests. Much more testing has been done using FEA computer analysis. A std book Locost chassis is around 2300 Nm/deg when computer analysed for instance. A Westfield would be similar given that a Locost is essentially the same thing. Computer analysis is great when designing chassis for 'what if' scenarios as it is much quicker and easier to virtually add a bit of cross bracing or chaning the chassis memeber sizes than doing it for real! Plus you also see the weight effect of making changes, though that's pretty easy to do on paper anyway.

The value for a DRB GT40 is 7250 Nm/deg, which again is realistic.

Manufacturers, as previously stated, don't tend to give out actual torsion results, but love the % increase reporting with each new model. I do remember reading that the new (at the time) BMW Z4 was three times torsionally stiffer than the old Z3 with a figure of 14500 Nm/deg. So the Z3 would be a little under 5000 Nm/deg one could infer. I also remember reading very recently a newish car was claimed to be around 30000 Nm/deg, which was apparently some sort of record for a production car. Wish I could remember what car it was though!

So, how to test torsional stiffness? Well, unlike the photo shown earlier, and also unlike the CAV GT40 test method, the load should be transferred to the chassis using the wheel hubs and suspension, the same as any real life load would be. I had trouble swallowing the amazing results for the CAV, but when you see how they tested it, well it's not comparing apples with apples now is it? Nice test setup though.

Have a look at the attached photo, which is actually of the +100 Locost mentioned above being tested. Note the chassis is fixed to ground via rear hubs. The torsional load is applied through the front hubs. If you're very observant you may note the front pivot is located on rollers also to take out any chance of loads building up due to restrained lateral movement under load. The other trick is to replace the spring/damper units with solid bars to prevent suspension movement. Dial gauges are located down both sides of the chassis and simple calculations allow conversion of vertical movement to angles. In this way the chassis siffness can be checked for abrupt discontinuities along it's length. Also a pair of dial gauges is located at the rear (fixed) axle line to subtract out the minor twist that will occur there even though it is 'locked' to the ground.

Think about what happens to a chassis due to suspension load inputs. In the case of double wishbone suspension (typical on front of most kitcars) a load travels up the spring/damper unit, but also loads are applied into the chassis through the wishbones. So, you could make a chassis really good at resisting forces through the damper mount, but forget about the wishbone mounts if you tested just applying loads to the damper mounts!

Anyways, there's a lot more that can be said, but it gets a bit 'engineering nerdy' and I can't be bothered! Lazy me...