R data.table: Difference between nested regressions results

Question

I am comparing two alternative strategies to estimate linear regression models on subsets of data using the data.table package for R. The two strategie