It seems to me that the structure of VMS 'words' (i.e. the characters between the spaces) clearly indicates that the spaces mean something and are not sprinkled in at random. The fact that there are places where the spaces seem to be 'out by one' suggests to me that the rule for inserting a space was not necessarily straightforward, and that sometimes the scribe forgot, or got it wrong.
The scheme I suggested some time ago (14/8/03) for multiple-table-driven encoding can explain the word structure and the placing of spaces. The rule for spaces would also have been a detail that had to be remembered when the table 'wrapped round', so it's quite likely that a space may have been forgotten or misplaced occasionally. In this scheme, the spaces do not represent word boundaries in the plaintext - which would have been another place where I can imagine an extraneous space being inserted accidentally (or possibly deliberately to avoid an ambiguity).
Cheers,
Jon.