Is there any mechanism then to at least make it unlikely that a transaction will "fall through the cracks" and never get processed?  Or is the idea that that's simply a risk you're taking if you skimp on transaction fees?

You'll keep resending it forever until it gets in a block.

Until your transaction gets into a block, nothing is guaranteed.  This is an untrusted, peer-to-peer network after all.  Bitcoin nodes try to remember your transaction, but the client must resend in case they do not.  New nodes appearing on the network have never seen your TX, and nodes may be restarted or leave the network, also forgetting your TX.

One of my proposals for bitcoin is to introduce a new rule that causes all nodes to forget a transaction, if it does not make it into a block within 24 hours (~150 blocks).

Clients that wish to resend for such transactions will continue to do so, submitting a copy again after 24 hours passes.  This is current behavior.

Clients that wish to cease retransmission, perhaps in order to rewrite the transaction into something more acceptable, are newly permitted to do so with my proposed change.

Thus, transaction lifetime would become deterministic.

Then it's probably 50 KB, I know that it's not much.
I believe that the fee structure is a condition of validity, but miners are not compelled to include free transactions at all.  In a future that there are enough paying transactions to max out the total blockspace, free transactions are likely to be ignored my profit minded miners.  Although there will likely be some miners that always include free transactions on ideological grounds, there is no way to know if free transactions will continue to be processed on any reasonable timescale.


Probably, but the 1 meg hard limit was an arbitrary rule included early on as a method to limit harm in the event of a 'transaction flood' spam-type attack.  It was never intended to remain in effect, and to my knowledge, could be removed without effect now that the transaction priority score rules and the minimum transfer fee rule are effective.  As far as I know, we have never even come close to the 1 meg hard limit anyway.   

The number of miners do not affect the block rate in the long term, because the system adjusts to changes in the network's total hashing power every 2016 blocks, or about two weeks.  However, within that two week window, it is expected that increasing transaction fees during peak periods will eventually create an "elastic" processing rate by drawing in miners who are marginal with just the block reward.  The idea being that during the high transaction periods, the accumulation of fees that are seen unprocessed on the network will trigger automatic generation to try to process those fee paying transactions, and the average time between blocks will drop.  The flipside being that on the weekends, and the off periods of the major Bitcoin trading areas, will end up having an average time between blocks longer than 10 minutes to compensate.

In short, "will there be enough miners?" is not an irrelevant question, even outside of the context of security.  However, the answer is always "yes" because the ebb and flow of the transactions and their average fees will create a fast moving market that will adapt to the conditions consistently.

IMHO that's brilliant.  Thanks for explaining!