The VM's are 32-bit - Dump generation is technically functioning, over the space of 2 days I managed to get roughly 150MB of work done on each of 2 VM's, but for some reason every few hours each VM was getting submitting a completed slice back to the system then getting a new job to reload one or both of the commons and en wikidumps, despite the commonsdate and wikidate files locally having the same dump version already stored (ie the dumpathome client was deciding to redownload and overwrite the existing local dumps with exactly the same freshly downloaded dump every few hours).
After doing so, the client then seems to have about a 50% chance of just carrying on processing fine, or start throwing out the errors listed above. Presumably something's amiss with having to redownload the wiki dumps so frequently in the first place, I can't see any real reason it would need redownloading at all, and having to redownload the 15GB ish of dumps twice a day is taking up more time than the VM's were able to actually spend processing (while both working they were averaging a slice every 10 minutes ish, between them).