This is right -- plus lm() is faster! Although, from a statistical perspective, if you can't invert X'X, that should first make you think "I have data quality issues" (i.e. multicollinearity) rather than "I need a different algorithm to compute the inverse".