Why are big single center trial effects not often replicated in larger multi-center randomized controlled trials?

Dechartres et al did a meta-meta-analysis on 48 meta-analyses comprising 421 randomized clinical trials. Across all their sensitivity analyses, reported treatment effects were consistently and substantially larger in single-center trials than in multi-center trials. No one knows why yet. Ann Intern Med 2011;155:39-51.