As You said, the most common situation for recommender system is to predict rating. Then RMSE/MAE is used. For results of a ranked item list different measures are used, e.g. Prec@K, Rec@K, AUC, NDCG, MRR, ERR. However, the are many algorithms for recommendation with its own hyper-parameters and specific use cases. Depending what You are trying to achieve You should benchmark Your solution with suitable method. The best source for this can be publication, where as You noticed, many times results for ML dataset are given. What's more about benchmarking recommender systems, sometimes the best prediction results are not so important as other aspect e.g. novelty or diversity.
For most common methods try this benchmark results for librec project:
There are no results for ML-10M, only ML-1M for rating and ML-100K for ranking. But You can always run chosen algorithm by yourself. There are other projects of this kind: http://www.mymedialite.net/, http://lenskit.org/.
Saying about evaluation of recommender systems this project should be mention:
It's dedicated for data preparation (splitting) and evaluation for recommender systems. It was presented at last RecSys'2014 . And also look at this . There are some benchmark results too.
 A. Said and A. Bellogín, “RiVal – A Toolkit to Foster Reproducibility in Recommender System Evaluation,” pp. 371–372, 2014.
 A. Said and A. Bellogín, “Comparative Recommender System Evaluation : Benchmarking Recommendation Frameworks,” pp. 129–136.