According to Vitalik Buterin, Ethereum should reduce the number of precompiles and instead implement EOF, followed by SIMD and EVMMAX. This approach aims to replace half of the existing precompiles with bytecode, making the system more future-proof. He highlights that this would be beneficial for projects like StarkWare, which require faster on-chain Poseidon computation but face challenges due to changing parameters. SIMD would allow such projects to manage these computations independently.