Looking at the validation report for model 6R7M, we see that there are no symmetry-related clashes listed. (Although there are two clashes within the asymmetric unit, HA of ASP20 - HD3 of PRO21 and HG13 of ILE95 - HB2 of ALA111)
The relevant portion of the PDB's validation explanation page is here:
All-atom contacts within the ASU are calculated by the Reduce and Probe programs within MolProbity (Word et al., 1999; Chen et al., 2010). This method was developed to quantify the detailed non-covalent fit of atomic interactions within or between molecules (H-bonds, favorable van der Waals, and steric clashes). Since most such interactions involve H atoms on one or both sides, all hydrogens must be present or added (Reduce optimizes rotation of OH, SH, NH3, etc. within H-bond networks, but methyls stay staggered). At present, in order to ensure comparable scores between NMR and X-ray, hydrogen atoms are removed from the analysed structure, and replaced by a different set placed by Reduce in idealised and optimized nuclear-H positions. All-atom unfavorable overlaps ≥0.4Å are then identified as clashes, using van der Waals radii tuned for the nuclear H positions suitable for NMR (rather than the electron-cloud H positions suitable for X-ray).).
(bolding mine)
The main takeaway is that MolProbity does not consider that amount of overlap to be a problem. So to your second bullet point, I don't think anything needs to be resolved.
As for which radii should be used, I don't have good advice. It looks like that is still being debated. However, looking up the original paper for the MolProbity tools used by the PDB for validation, the radii for O and non-carbonyl C are 1.4 and 1.75 Angströms, respectively. That sums to 3.15, I measure 2.92 between the atomic centers, so slightly less overlap than you got.
My initial assumption was that these would be poorly-resolved residues near the edge of the macromolecule and that the VDW overlap would be due to fitting errors, but it looks pretty good to me:
As noted in comments, some nonbonded atoms in model 6R7M have more than 0.4 Angströms overlap, but are not tagged as clashing, e.g. the sidechain oxygens of THR29 and TYR73. This is because these overlaps are not unfavorable; that is, molprobity predicts hydrogen bonding between them. I couldn't find any specifics about how molprobity determines if a particular overlap is unfavorable. The best description I found is in the documentation of the chimera tool findclash:
The Clash/Contact Parameters control what will be considered a clash or contact. The overlap between two atoms is defined as the sum of their VDW radii minus the distance between them and minus an allowance for potentially hydrogen-bonded pairs:
overlapij = rVDWi + rVDWj – dij – allowanceij
Find atoms with VDW overlap >=[ cutoff ] angstroms - pairs of atoms with overlap ≥ cutoff will be identified. A larger positive cutoff restricts the results to more severe clashes, whereas a negative cutoff can also identify favorable contacts.
Subtract [ allowance ] from overlap for potentially H-bonding pairs - an allowance > 0 reflects the observation that atoms sharing a hydrogen bond can come closer to each other than would be expected from their VDW radii. The allowance is only subtracted for pairs comprised of a donor (or donor-borne hydrogen) and an acceptor. This is equivalent to using smaller radii to characterize hydrogen-bonding interactions (for example, see Li and Nussinov, Proteins 32:111 (1998)). As in FindHBond, possible donor groups are hydrogen-bearing nitrogen, oxygen, and sulfur atoms, and possible acceptor groups are nitrogen, oxygen, and sulfur atoms with a lone pair.
