Many files begin with a "well-known" header sequence which I imagine helps a lot in performing cryptanalysis over a huge number of data transfers.
So I have the thought of prepending a block or two of random data to make cryptanalysis more difficult.
Let's say that the file is like this:
<Header 250 bytes><Contents>
So we prepend some random data (of random length) like this:
<Random data of random length><Magic Number><Header 250 bytes><Contents>
The receiving side, after decrypting, simply scans for
<Magic Number> and snips everything up to that point.
Since the length is random, the attacker won't be able to guess where exactly the well-known parts will be.
Do you think this is a good idea? Or too much effort for not enough improvement?