For a consensus algorithm based on proof-of-work, the threat of an optimized closed-source miner is non-negligible, especially if it comes to altered or new hash algorithms that claim to be ASIC or GPU resistant.
As part of my community work on CryptoNote-based forks, I helped optimize the existing implementation, fix an exploit of the hash algorithm, and provide a proof of concept implementation using OpenCL.
As a result, all miners had access to the faster implementation, and the exploit got fixed, providing more security to potential 50 percent attacks.