Not to intrude into this marvelous conversation, but... RJ, while I generally agree with some of what you've said, it breaks down a bit when you include mtDNA into the conversation.
Our friendly little organelle contains a grand total of 16,569 base pairs in its DNA molecule. Not millions, thousands. As DNA goes, the whole thing isn't even large enough to register as a single segment if we were to plop in down onto, say, chromosome 8 and look for it with our current direct-to-consumer autosomal DNA testing technology.
Included in those scant 16.5K base pairs are a single regulatory region and 37 genes that code for 13 polypeptides, 22 tRNAs, and two rRNAs. These account for over 80% of the 16.5K base pairs. Mutation can't mess much with that regulatory region and those 37 genes without causing a big ouchy to the host human cells.
That's why, after the mtDNA genome was sequenced, initial testing of mtDNA for population genetics was limited to tiny areas at the very start and the very end of its circular structure. The non-coding Hypervariable Regions were so called because they could be, well, variable: free to change without botching up some really important piece of transfer RNA, for example. Now, why the Hypervariable Region at the beginning of the mtDNA structure is called HVR2, and the one at the end of the structure is called HVR1...well, there have to be some mysteries, I suppose.
Anywho... Only a little over 3,000 base pairs in mtDNA are not directly involved in the regulatory region and the 37 do-important-stuff genes. HVR1 and HVR2 account for 1,144 of those.
In aggregate, the parts of the mtDNA genome that can safely mutate and form novel SNPs is ridiculously small. And it's the same mtDNA that's been with us since we crouched in the Great Rift Valley in Africa and marveled at having opposable thumbs...never dreaming that the anatomical oddity would lead to an obsession with using smartphones for everything.
The sparsity of genetic material (as well as heteroplasmic frequency and the sheer number of the little buggers in our bodies, on the order of 595.2 trillion) is one reason that I've never considered mtDNA very fertile ground for genealogy. Can be a useful tool to disprove an hypothesis, but as a means in and of itself to confirm a genetic match it's quite weak. Still only four nitrogen bases, and adenine still bonds with guanine, cytosine with thymine. The world population clock says there are 7.6 billion of us alive right now. There simply isn't enough possible variability in the tiny mitochondrial DNA molecule to differentiate to any genealogically granular level among the population. Your mother and my mother might be an exact match out through the coding region, but the MRCA might not be closer to us than 20 or even 50 generations ago.
Which brings me roundabout to the notion that there were any--much less many--mtDNA SNP variances which existed in prehistory but that have gone extinct. I'd argue that the mtDNA molecule simply doesn't have enough genetic material. Going by the numbers only, my guess is that we have more unique mtDNA genomes in existence today than ever before. If we can believe the paleoanthropologists, at the time of the supposed Toba supereruption 70,000 years ago, we were down to as few as 10,000 humans. In the early Holocene, after the last glacial period about 10,000 years ago, there were between one and ten million humans (with an uncertainty of up to an order of magnitude).
So--this is gonna make all two-dozen of my cerebral neurons hurt--there would have been some 5.95x1021 mitochondria in existence and replicating 10,000 years ago. Today there would be 4.52x1024. The population of human mitochondria then would have been only 0.13% of what it is now. Seems to stand to reason that we're likely to see more mtDNA mutational possibilities in the much larger population. Something I should probably try to research before I comment, but it's Saturday morning and more coffee is calling: I also can't recall ever seeing a study of human mtDNA taken from ancient remains that was thought to be of an extinct haplogroup, no longer present in the population. If anyone knows of such a study, please point me to it. Save me the research. :-)