Also, what is the point in wasting fees by transacting over wallets with the same owner? Unless you prove metcalfe's law wrong the only achievement of such waste is to keep the price (value) of bitcoin high, so i take it you mean it's a speculative manipulation.
Even so, with a true wallet-to-many scenario (OpenBazaar) the percentage you are proposing here are doomed to shift.
Two obvious reasons why someone would want to move bitcoins between wallets of his own are (1) tumbling and (2) hotwallet/coldwallet management. Someone may also be (3) torture-testing wallet software. (All the blockchain traffic could easily be created by a single person with a modest BTC capital.)
Last time I checked, very few transactions included fees. Right now the mining network is financed by "printing" new bitcoins, to the tune of 10% of the existing bitcoins per year. (That is, 10% yearly inflation, in the strict sense of the term...) Since address creation and transactions are free, there is nothing to discourage fake blockchain traffic.
Fake traffic may include also (4) intentional attempt to inflate the transaction volume. Right now, if the cost of the mining network were to be paid by fees rather than "inflation tax", the fee would have to be nearly 4% of the total transaction amount. But if fees were charged then the "fake" volume would all but disappear. If only 50% of the transactions are real e-payments, then the cost of processing one transaction would be 8% of its amount; if only 10% is real, the cost would be 40%. Thus someone may be inflating the volume to hide this unpleasant fact.
Some blockchain traffic is also (5) people depositing and withdrawing BTC from exchanges and other services with individual accounts. Although that traffic is not "fake" by my defintion, it is still sort of fake because the BTC on the exchange still belong to the client, logically.
(Bitcoins changing accounts inside the exchanged do not generate blockchain traffic, but they are not real e-payments either, since there is no counterpart transfer of goods or services -- merely a swap of one currency for another.)
Metcalfe's law seems to hold when the quantities are plotted in logscale over the last 5 years. However, the last year is compressed to a tiny area at the top right corner of that plot. At that scale, a deviation of 90% would hardly be noticed, since it would span less than 20% of the vertical scale. If the data is plotted only over the last year, the fit is not that good. And, anyway, if the traffic had been 90% "fake" over the last 5 years, Metcalfe's law would hold just as well.
