Patriot looks at a variety of run estimators using basic data (excluding SF, SH , GIDP etc). The results aren't particularly surprising: on a team-seasonal basis, all the estimators are in the same ballpark (in terms of RMSE); Base Runs is not any less accurate and is perhaps a little more accurate on a team-seasonal basis.
Tangotiger's comments are contained in the Primate Studies section at Baseball Primer. He suggested creating a best-fit formula for all odd-numbered years (sample A) for each method and then testing it on the even-numbered years (sample B). The reasoning there was that if a formula was really vaild on a team-seasonal basis, the loss of accuracy when going from the "best-fitted" sample to a completely different sample would not be great. Patriot tested Base Runs against the linear best-fit model for sample A and found the former to be slightly more accurate on sample B.
Here are some other articles from Patriot's handy Sabermetrics section:
Patriot's introduction to run estimators. He briefly discusses the two main non-linear "competitors": Runs Created and Base Runs.
He explores some of the ways one can utilize Base Runs. From that page, you can click on the link: Base Runs: A Promising New Run Estimator for Brandon Heipp's succinct overview of Base Runs.
Another of the interesting links is Patriot's analysis of EqR
https://www.battersbox.ca/article.php?story=20030912011008999