I know and like RC4 and while I was aware of the weak-keying issue, that can be mitigated when choosing the initialization vector (or if using the algorithm as a proper stream cipher, rather than a block cipher as it was in WEP).
The "L shaped pattern" described in the PDF is a new one on me, even though it was apparently described in the same paper 13 years ago. Not sure it's a home run however, it still relies on a (larger) class of weak key, and (if I've read that paper right) the best case is those are only 1 in 2^16 of sessions. So an individual RC4 encryption is likely still fine, it's only when there are millions of them that one becomes statistically likely to fail. The odds of that one having a password, credit card or whatever are still low.
So I'm not going to panic just yet.