From a benchmark point of view, your model needs a refinement regarding processor cache hits / misses ( computational aspect ) and filesystem cache hits / misses ( I/O aspect ). If possible, you want to eliminate these aspects.
Because your impression was that encryption / decryption is relatively expensive (see your 1st post), I would like to suggest you to do the following experiment. Moreover, because ehab wrote:
"on what base you suggested :
No, encryption is very fast (well, it should be), including AES "
I also would like to invite him to do the same experiment :
Since you're both truecrypt users, start Truecrypt => Tools => Benchmark...
And let us know the AES results, please specify your processor type, GHz, #cores, etc.
I already mentioned a 190 MB / s TC result, 2.4 GHz duo core, somewhere in another discussion ( this forum ). In case my "should be" remark doesn't hold for BC somehow, that's of course easy to solve.
I still think a number of other BC benchmarks / changes may be more useful.
If you disagree, let me know ;-)