Compression will screw up character frequencies [...] enough to make automated detection of a successfully-broken encryption really darn hard.
The question is just how hard is "really darn hard"? Compressed English text has characteristic patterns just as plain English does. The salient difference is that these patterns take longer to emerge at the same confidence level. The compressibility limit is a limit not usually reached; the difference between that limit and the actual compressed text will be non-zero. This difference manifests itself in patterns in the compressed text. Some estimates of this size are necessary in order that the designer have an assurance that automatic recognition of decrypted text is difficult. These concerns are largely obviated by using ciphers with longer key lengths, of course. Eric