I saw Dr. Lolle present this research at an international conference, so I can give a few more tantalizing details:
Plants from a strain that had been bred to be homozygous for a maladaptive allele at a certain locus (and were so, beyond a douct) somehow managed to produce offspring with the adaptive wild-type allele that hadn't been in their lineage's genomes for several generations. That's completely non-Mendelian inheritance: a cross of hh x hh could yield Hh or even HH. This means that the old allele had somehow been stored and passed on for generations somewhere outside the DNA.
The team speculated that there must be a previously undiscovered RNA-based template that stores old copies of genes for several generations, and causes DNA to revert if the plant isn't doing well. It would be a cache of genetic variation stored and inherited entirely outside DNA.
It seems certain that something like that must be happening, so for the sake of argument, assume that's exactly what's going on: there's an additional heritable place where genetic information can be backed up, but it's only used when called upon. Furthermore, assume that this mechanism exists in other eukaryotes (e.g. humans). What are the applications to genetic therapy and engineering?