Rough numbers: At 300GH/s and difficulty of 1, I should in theory mine one valid block in 3ps (picoseconds), meanwhile at 6MHz it would take 166ns (nanoseconds).
However, in a difficulty=1 scenario, the difference of 165997ps should make no difference at all. Network latency to other nodes will probably be in the single to double-digit ms range. For reference: 1ms (which is a really good network latency) are a million ns.

There is a chance that your CPU is overloaded with all cores mining and no core ready to push that packet out to the network asap. Another idea would be making sure that this old CPU miner code doesn't wait for e.g. a whole set of nonce values before returning a valid block or something silly like that. It should return as soon as a block meets the difficulty, but I wouldn't rule out that it waits longer for whatever (legacy?) reason.

I will definitely have a look at your bfgminer config and the bfgminer code, too.