Collecting historical phoneme error rates for the TIMIT corpus and error rates for the winning entreis in the annual Imagenet Large Scale Visual Recognition Challenge. The data is collected on this spreadsheet. It is currently completely open, so feel free to improve or add papers and error rates.
The data is presented on this blogpost