<p>Well, the "security" of the rolls mostly depends on the attacker not knowing how it works, and the majority of the population being comprised of illiterate slaves. Having 3 or 4 rolls of different sizes would suffice for cracking most of this; knowing the fact of variable roll size is 99% of the trick.</p>
<p>A considerable portion of security from encryption is achieved by minimizing the source text, notably to keep out readily guessable plain text. In WWII, British decryption efforts were helped plenty by German officers that considered it a "career limiting" factor *not* to end messages with "Heil Hitler". That perception mayn't have been wrong, but those bits of predictable plain text almost certainly caused the losses of German U-Boats, as it provided a vulnerability for Allied cryptographers to exploit.</p>
<p>F. L. Bauer's book on cryptography describes other "politically necessary" sorts of cryptographic protocol failures - when messages contain fawning phrasing ("by order of the fuhrer") or spelling out some of the wacky long officer titles in high command, this all helps in attacking:<br>
A) individual messages<br>
B) message keys that will be used on other messages<br>
(Hence, the sloppy bozo may wind up getting others that are competent killed)<br>
C) the cipher system as a whole.</p>
<p>Bauer observes that a *good* cryptograms clerk:<br>
- removes all unnecessary text<br>
- abbreviates heavily<br>
- misspells whatever they can</p>
<p>That seems like it's likely to still be valid-ish.</p>
<p>We do have stronger ciphers, today, but the notion that having known plaintext helps certainly persists in modern cryptanalysis. You'll see it a fair bit in Bruce Schneier's writing (sp?)</p>
<p>On some extra reflection, there is a harmful aspect to encrypting your whole system, as this introduces a barrel load of known plaintext. Forget about a few references to Nazi haute, you are throwing in a dozen copies of the GPL, and as likely as not, a gigabyte of well-known binary and text data. Lots of material for cryptanalysis, quite possibly enough to meaningfully enhance a brute force attack.</p>
<p>And when it's certain that the key for all that will be in the VM, a smart attacker won't bother with brute force when getting the key from the VM will provide the Keys To The Kingdom. Better still, once cracked, you can't fix it - changing the key requires rebuilding your VM. A *really* smart attacker may be sufficiently ready that they'll regain access before you can reboot into the new VM!</p>