I'm not sure it would interact nicely with the request authentication system that S3 uses.
I think a generic cache layer would be a better solution. A bit of googling turns up fuse-cache, which sounds like about the right thing (although I haven't actually examined it in detail), or fs-cache, which also sounds like a discrete cache layer to be added to any file system. Basically, you mount it, and it passes requests through to any other mount (like s3fslite) while adding an on-disk cache layer.
I haven't tested any of these, but that seems like an approach worth pursuing.
I'm hesitant to automatically fire off that many requests, especially when they may not end up being necessary. If you are using the same machine and preserving the cache, it will already be primed each time you mount the bucket, except the first time (or any time you delete the cache database file).
Using find is just a trick I used whenever I'd corrupt the cache or change the DB schema while developing it, and then wanted to go in and test it again interactively.
I should probably mention that reducing the number of requests was one of my primary goals. The first time I played with s3fs (the one I forked), my bill for the month was roughly 10% storage and bandwidth, and 90% requests (or was it 20/80?).
Anyway, thanks for the feedback; I do appreciate it!
I do pay money to subscribe to a newspaper that pretty much fits your description (The Economist). I would amend your list of qualities to say that the viewpoint of each article is always stated and argued. Most sources claim to reach no conclusion, but they actually do. The honest approach is to argue a point and try to convince the reader that you are right, not try to manipulate the reader into agreeing with you unknowingly.
Read any of the Jon Stokes articles, for example, and he usually makes predictions or makes a claim about the hidden reasons behind some industry action, but he doesn't pretend otherwise. He states that he thinks x was a bad move, and then explains why.
I thought maybe it was just me. When I was a postgrad at Cambridge (2003-2006) I supervised for this course a few times, thinking that a course surveying a bunch of interesting languages would be fun. I seem to recall a bit of smalltalk thrown in, but it was mostly BCPL. I walked by Martin Richards' office and noticed a poster announcing a port of BCPL to .NET or some similar abomination.