My personal priority are similar to Vincent's. If I cause a regression, I try to fix it
*) If its an easy fix, useful hints in the bug report etc, and my time permits *) Or if the fix is important to the one who pays my electricity bills(CodeWeavers) *) If the regression is a real regression because the patch had a flaw, and not because it happened to uncover existing other bugs(for example, I do NOT consider Bug 18401 a regression) *) If the regression happens in an area of code that is generally shaky and needs more attention to fix. In that case just trying to patch out the regression will cause new regressions - directly working on the big fix as time permits is usually better.
There are surely some exceptions to this etc and I guess I left some easier regressions I caused untouched because I simply lost track of them. Every now and then they get randomly fixed or someone else picks them up.