Our testing showed us that if a robot doesn't clear balls first, there will be 2 smashed-together balls between the bumpers and the wall nearly all of the time. That was about 6" of gap.
If a gear was in the mix, the smallest we could make the gap was 11" using grippy-ish wheels on a 4-CIM 110lb testbed. That's about 14" to the frame perimeter...
The gap itself isn't a big deal since the excess balls & gear are probably cleared just by the incoming pathing* - thus there's likely to only be 1-deep worth of smushed balls creating a 2-3" gap. The fact that the robot can't use the wall as an alignment device if there's an unpredictable gap is a pretty big deal for some of the more simplistic designs (e.g. Ri3D & designs that can be easy Cheesecake). The WCP MCC concept deals with the gap the best, but teams will still want to work to improve upon it.
It will be interesting to re-test the gaps using these new drawings though. I forgot to account for the corners

.
*It's kind of a big deal - the easiest path to the gear slots that clears balls basically pushed the balls closer to the opponents' boiler...