What happens if bitcoin blocks fill? Miners choose transactions with the highest fees, so low fee transactions get left behind. Let's look at what makes up blocks today, to try to figure out which transactions will get "crowded out" at various thresholds.
Some assumptions need to be made here: we can't automatically tell the difference between me taking a $1000 output and paying you 1c, and me paying you $999.99 and sending myself the 1c change. So my first attempt was very conservative: only look at transactions with two or more outputs which were under the given thresholds (I used a nice round $200 / BTC price throughout, for simplicity).
(Note: I used bitcoin-iterate to pull out transaction data, and rebuild blocks without certain transactions; you can reproduce the csv files in the blocksize-stats directory if you want).
Paying More Than 1 Person Under $1 (< 500000 Satoshi)
Here's the result (against the current blocksize):
Let's zoom in to the interesting part, first, since there's very little difference before 220,000 (February 2013). You can see that only about 18% of transactions are sending less than $1 and getting less than $1 in change:
Paying Anyone Under 1c, 10c, $1
The above graph doesn't capture the case where I have $100 and send you 1c. If we eliminate any transaction which has any output less than various thresholds, we'll catch that. The downside is that we capture the "sending myself tiny change" case, but I'd expect that to be rarer:
This eliminates far more transactions. We can see only 2.5% of the block size is taken by transactions with 1c outputs (the dark red line following the block "current blocks" line), but the green line shows about 20% of the block used for 10c transactions. And about 45% of the block is transactions moving $1 or less.
Interpretation: Hard Landing Unlikely, But Microtransactions Lose
If the block size doesn't increase (or doesn't increase in time): we'll see transactions get slower, and fees become the significant factor in whether your transaction gets processed quickly. People will change behaviour: I'm not going to spend 20c to send you 50c!
Because block finding is highly variable and many miners are capping blocks at 750k, we see backlogs at times already; these bursts will happen with increasing frequency from now on. This will put pressure on Satoshdice and similar services, who will be highly incentivized to use StrawPay or roll their own channel mechanism for off-blockchain microtransactions.
I'd like to know what timescale this happens on, but the graph shows that we grow (and occasionally shrink) in bursts. A logarithmic graph prepared by Peter R of bitcointalk.org suggests that we hit 1M mid-2016 or so; expect fee pressure to bend that graph downwards soon.
The bad news is that even if fees hit (say) 25c and that prevents all the sub-$1 transactions, we only double our capacity, giving us perhaps another 18 months. (At that point miners are earning $1000 from transaction fees as well as $5000 (@ $200/BTC) from block reward, which is nice for them I guess.)
My Best Guess: Larger Blocks Desirable Within 2 Years, Needed by 3
Personally I think 5c is a reasonable transaction fee, but I'd prefer not to see it until we have decentralized off-chain alternatives. I'd be pretty uncomfortable with a 25c fee unless the Lightning Network was so ubiquitous that I only needed to pay it twice a year. Higher than that would have me reaching for my credit card to charge my Lightning Network account :)
Disclaimer: I Work For BlockStream, on Lightning Networks
Lightning Networks are a marathon, not a sprint. The development timeframes in my head are even vaguer than the guesses above. I hope it's part of the eventual answer, but it's not the bandaid we're looking for. I wish it were different, but we're going to need other things in the mean time.
I hope this provided useful facts, whatever your opinions.