Whilst I can understand Google's desire to parallelise and accelerate DNS resolution, building their own browser-level resolver is a hairy way to do so.

Several projects have attempted to address this problem before ( C-ARES, and Polipo's internal resolver ) but never managed to achieve critical momentum. Might it not have been worth focusing effort there?

Plus, naturally, there's the temptation to use Google's own public DNS servers instead of adhering to the OS-specified resolution order.