From Simon Johnson:

I have been working with PUBLISH -1 in v2. I have found a flaw with the new packet type layouts. Presently you can PUBLISH -1 without a session, and it is the SESSION that determines the protocol version you have decided to use. Without a session, the GATEWAY has no mechanism to determine the version of the PUBLISH packet to use to decode it. So to maintain the ability for a GW to support both protocol versions, we must have a mechanism in the PUBLISH packet to note protocol version WITHOUT adding any further bytes.

I was looking, the packet type field seems to be a great chance to squeeze some more space out of - in a similar fashion to tt5, so we coolocate a generic flags field in bits 1 & 0, using 6 bits of space (7-2) for packet type (64 available types - we currently use ~30) so plenty of growing room - then the 2 new flags can be used for flags on the message without needing a new flags field. In the case of PUBLISH this can be protocol version, but also it could be used for a few other types specifying and using only 1 or 2 flags saving space elsewhere. I feel this is a fundamental enough issue to warrant this change - and give us further flexibility for saving space elsewhere.