I agree - or rather a friendly chemist agrees - that the structure is a bit rubbish. The first kekulize misassigns the double bonds in the carbon you mentioned, so the second aromatic assignment does not regard these bonds as aromatic, then the second kekulize does not operate on the same structure as the first. I agree that this is not ideal behaviour, but it is comprehensible.
The only solution I can think of is to assign all bond types:
where the double bond to the phosphorus is detected, the five membered ring is no longer aromatic and the kekulisation works as expected.
I have mailed the database group to see if they want to fix the bonds in the structure, but this will be too late for the forthcoming November release.