That is also incorrect. 2 inputs from the same address are exactly the same size as 2 inputs from 2 addresses. Bitcoin doesn't work on addresses and balances it works on inputs and outputs. Using new addresses for each transaction doesn't make transactions even a single byte larger.
* To hit 7 tps with 1MB blocks would require the average txn to be 238 bytes. A P2PkH txn with compressed keys (and sadly more than 30% of txn today still use uncompressed keys making them much larger) is going to be 10 bytes + 147 bytes per input + 35 bytes per output. So even a 2 in, 2 out txn with all compressed keys is 374 bytes. To hit 7 tps (238 bytes per txn) would require 100% of txns to consist of nothing but 1 in, 2 out txns with compressed keys (227 bytes ea). So realistically even not considering people who use uncompressed keys, OP_RETURN, using multisig, P2SH, etc the realistic limit is closer to 4 tps. 2015 will probably be the year of multisig and complex txns so I think planning for larger txn sizes makes sense. The network is unlikely to reach 3 tps without a larger block limit.
Here's a chart that uses extrapolation to estimate the dates when the bitcoin network would reach 3 tps and 7 tps if historical growth rates continue to hold. If we witness another "growth spurt," I expect we will approach 3 tps on several days and the discussions with regards to increasing the blocksize will become more focussed.