That being said, this is testable ...

If you post a little (but self-contained) snippet that shows how you tested that writes are being done in 4k blocks, I might run it with my 8k buffer perl I compiled back then, which I still have around.

Update: ...unfortunately, I couldn't get Fuse to work (with a reasonable amount of effort) on my somewhat older SUSE system at work — module builds fine, but tests fail, although kernel-wise, etc. fuse is in principle working.  I'll see what I can do later at home on my Ubuntu system.

The problem with posting it is that there is rather a lot, and Fuse requires a lot of implemented subroutines for writing (reads, exists, unique and so on) so if you don't mind it'd be a lot easier for me to just give you a tarball (http://www.jamescondron.co.uk/steg.tar.bz2) of the entire implementation (8k).
To test I used dd and watched the debug output from the script (fs_simple.pl). Invocation:


perl ./fs_simple.pl <mountpoint>
[download]

I've played around with this some more (on Ubuntu), but I'm not really getting anywhere. I.e., with my perl-5.10.1 I somehow can't get the Fuse module to work:

~/PM/build/Fuse-0.09$ make test
PERL_DL_NONLAZY=1 /usr/local/bin/perl "-Iblib/lib" "-Iblib/arch" test.
+pl
test/s/mount.t ... 1/3 # mounting examples/loopback_t.pl to /tmp/fusem
+nt-almut
# Mounted in 0.2 secs
test/s/mount.t ... ok   
test/chmod.t ..... sh: cannot create file: Function not implemented
...
...
Failed 7/21 test programs. 45/156 subtests failed.
make: *** [test_dynamic] Error 255
[download]

And I don't really know why. Doing exactly the same steps with perl-5.8.8, all tests succeed (but this is of no help here, as it's the 5.10.1 that has the 8k buffer...).  Also, my 5.10.1 is generally working fine (all tests did succeed), as is my system's fuse.

I think I'll give up for now... :(  Sorry.