#19426 introduced a noticeable slow down and memory increase of the linear models with dense input, as can be seen here https://scikit-learn.org/scikit-learn-benchmarks/

ping @ogrisel @agramfort @maikia