In the original Usenet thread about RC4 when it was finally reverse engineered, a user by the name of Michael Johnson mentioned that the RC4 algorithm looked surprisingly like another proprietary cipher which he had seen which he called "brand X", and mentioned several similarities between the two:
I am suprised at (1) how simple the cipher is, and (2) how closely it resembles another proprietary (but unclassified) cipher that I have seen. The biggest difference between alledged RC-4 and brand X is that brand X includes some data dependencies. Brand X is stronger against a plain text attack, and can be used for data authentication. The plain text attack part is not a big deal if you heed the caution in observation 1, above. Brand X performs 3 swaps per byte emitted instead of 1, and uses different index manipulation, but it also uses a 256 byte array and very similar principles of operation. The 3 swaps are not needed to avoid short stream cycles, but they do make the authentication properties work.
The key expansion of alledged RC-4 and brand X work a little differently. The alledged RC-4 version is faster than brand X, but is more likely to generate the same state from a different user key. The key AAAAA is the same as the key AAAAAAA. The key setup in alledged RC-4 makes no attempt to differentiate between different sized keys, but it does in brand X.
What cipher is he talking about? Does this description match any known cipher?