DOTS Physics Bevel Radius Performance Testing

There isn't a lot of information out there about the Bevel Radius option on DOTS Physics Shape components, so I figured I'd do my own experimentation and post the results here.

Here's what the documentation has to say about it:
"To improve the performance and robustness of collision detection Unity Physics allows contacts to be generated within a small tolerance of the surface of a convex shape. This tolerance is referred to as convex radius or bevel radius."

So it sounds like Bevel Radius improves performance, but adding a bevel to a box collider makes the geometry more complex, so which ends up actually being more performant?
Is more bevel radius always good, or does the performance start to drop off if it goes too high?

I set up a simple test to try to answer these questions.
The test consists of:
A wall of 3,680 stacked DOTS physics cubes falling onto a 500x500 terrain with 513x513 heightmap with some bumps on it (To attempt to replicate a real world use case with things interacting with a more complex geometry than just a box collider.) and an FPS script to track the lowest and average FPS while the tower collapses.

Here are the results:

0 Bevel Radius:
Test 1:
Low: 104
Avg: 209

Test 2:
Low: 97
Avg: 203

Test 3:
Low: 104
Avg: 220

Test 4:
Low: 104
Avg: 220

Test 5:
Low: 105
Avg: 211

.05 Bevel Radius (Default Value)
Test 1:
Low: 108
Avg: 219

Test 2:
Low: 108
Avg: 218

Test 3:
Low: 109
Avg: 219

Test 4:
Low: 106
Avg: 217

Test 5:
Low: 107
Avg: 222

.1 Bevel Radius
Test 1:
Low: 90 (Ouch, an outlier)
Avg: 219

Test 2:
Low: 114
Avg: 225

Test 3:
Low: 112
Avg: 220

Test 4:
Low: 110
Avg: 221

Test 5:
Low: 110
Avg: 218

.2 Bevel Radius
Test 1:
Low: 93
Avg: 223

Test 2:
Low: 89
Avg: 225

Test 3:
Low: 105
Avg: 213

Test 4:
Low: 102
Avg: 217

Test 5:
Low: 95
Avg: 218

.3 Bevel Radius
Test 1:
Low: 147
Avg: 218

Test 2:
Low: 111
Avg: 221

Test 3:
Low: 102
Avg: 214

Test 4:
Low: 112
Avg: 211

Test 5:
Low: 113
Avg: 223

Past .3 Bevel Radius the collider starts to be more of a sphere than a cube, so I figured I'd stop here.

Averages:

No bevel:
Average Low between 5 tests: 102.8
Average between 5 tests: 212.6

.05 Bevel (Default)
Average Low between 5 tests: 107.6
Average between 5 tests: 219

.1 Bevel
Average Low between 5 tests: 107.2
Average between 5 tests: 220.6

.2 Bevel
Average Low between 5 tests: 96.8
Average between 5 tests: 219.2

.3 Bevel
Average Low between 5 tests: 117
Average between 5 tests: 217.4

Conclusion:
I think the obvious conclusion is that any difference is very tiny, and if I were to repeat the tests the results could easily swing in the other direction.
However, my personal interpretation of the results is:
1: Having some bevel radius at all seems to marginally improve performance over having no bevel radius at all. (confirming what the documentation says)
2: Increasing bevel radius past the default of .05 might help an extremely tiny amount, but it seems like the majority of the gains are from just having any amount of bevel radius.
So .05 is marginally better than 0.
.3 might be extremely marginally better than .05, or at the very least definitely isn't worse.

Hope this is helpful for other people out there that like to min-max their performance.