The previous version worked well during ionization phases of a calculation, but was poor during recombination phases with few/no sources. The updated approach should work well for both.
The updated method could have a minor affect on test problems, but all should still be reasonably close to the previous solutions. It passed all of my local tests, so I don't anticipate any issues when it's tested in Jenkins.
This is expected, since the update will affect how adaptive time stepping occurs inside the FLD solver. This difference is very small, so I'd say that they should be equivalent, or that the new version should become the gold standard.