I feel like a DarkWallet should require the full chain, although I know that's an unreasonable expectation.
The actual BIP37 standard, and existing implementations of it, have a number of other flaws that reduce privacy. For instance the standard lets the seed value of the hash function be tweaked with a 32-bit integer, nTweak. However on the one hand if randomly chosen and rarely changed, as suggested by BIP37, the 32-bit integer can be used by an attacker to correlate multiple connections from the same wallet. On the other hand if nTweak is changed an attacker that can link multiple bloom filters can AND those filters together to greatly decrease the
false-positive rate and determine exactly what funds are in the user's wallet.
http://article.gmane.org/gmane.comp.bitcoin.devel/3625