Today I have provided a temporary work-a-round in the comparison method that resolves these rounding issues. So long as the rounding errors are in amounts that are effectively 0 economic value (1 or 2 satoshis) and greater than the minimum transaction fee this would not be a viable means of attack. The total 'inflation' from rounding would be limited to 4 BTC / year if this were bitcoin and all rounding errors went were positive and all transactions had rounding errors. Given that 90% of transactions have no rounding errors and 50% of the transactions round the other way, the net effect should average 0. With a little careful work I can always round down so that these rounding errors act as a kind of extra 'fee' the network charges users and no new coin ever gets created.

If using rationals was an option I could simply switch to 128 bit representation as it would have identical storage requirements and move the rounding errors so that they are so small as to not even matter.