I'm pretty sure that Crashfrog is wrong, because if it worked that way an indel IN the intron would affect the gene, so overall the risk of frameshift mutations would go up. The statistics don't work out in a way that favours his view, either.
If, on the other hand introns reset the reading frame, then the effect of indels would be limited to the start of the next intron - assuming that the frameshift doesn't mess that up. I'm speculating here, but it sounds like something that might be true....