Indeed. Which is why Matt's relay system and IBLT are needed, and will massively improve the situation for block propagation. People arguing for 1MB forever ignore this, yet when 20MB blocks finally appear they will take less than 1MB to broadcast through the network. It will only be bootstrapping new nodes and re-sync which will need the full blocks transmitted.
Those things help, but all full nodes still need to download all new transactions. Even with all proposed optimizations, you'll still need to download about 20 MB and upload 20*x MB (where x is some number that depends on but is probably less than the number of peers you have) for every 20 MB block you receive. This is better than the current network, which probably requires more than twice as much bandwidth for both download and upload. But it's not a magic bullet.
You proposed increasing the limit to 2MB in 2 years. This is what should have been done in Feb 2013 when this debate last came up in earnest. Unfortunately, this is too little too late. Of course IBLT was not considered at that time.
If 2 MB is all that can be agreed on now, then it's better to schedule that hard fork ASAP rather than agree on something marginally higher in 6-12 months. I personally think that 10 MB would probably be fine, though I know that there are many people who would disagree with me.