Speedup in well optimized cryptographic functions are hard to come by. In libsecp256k1 we'll usually celebrate a 4% algorithmic speedup. Figures on the order of 2x are reasonable for the usage in Bitcoin, though larger might be possible in the future during initial block download since much larger batches could be used.
A single validation on a single core of a fast desktop grade cpu takes about 50 microseconds.
Another relevant Q/A is Schnorr's batch validation