Very interesting stuff! While still thinking about all this, and reading around it, I would like to contribute to the discussion about why viruses sometimes kill their hosts.
It depends on a number of factors and dynamics. Viruses are obligate genetic parasites. They form two different levels of partnership with hosts. In acute viral infections, virus and host will often have a relatively fleeting level of interaction, with selection operating at mainly selfish level in both virus and host, and here there will be the real potential for viruses killing some of the host population. If the dynamics of viral evolution favour rapid spread, this can maximise disease manifestations, eg coughing, sneezing, infectious skin lesions, etc, and the evolutionary drive towards maximising symptoms in host can lead to lethality. Nowhere is this more dangerous to host than in the initial interaction between a highly infectious virus and a virgin host -- a so-called emerging virus infection. This is a situation where the virus might even cause the extinction of the virgin host. But one needs to take a step backwards and really think it through to realise that in such circumstances the virus itself may not be threatened with extinction in a broader sense.
A virus invading a virgin host will probably have hopped species (even families) to do so. In other words it is coming from another host with which it may well have established a long-term partnership.
Let us take the squirrel pox virus that is currently threatening the UK red squirrel with extinction. This virus is a persistent partner with the grey squirrel. In persistent infections -- partnerships where the virus never leaves the host (think myxomatosis, hantaviruses, AIDS, etc) -- the partnership dynamics are very different. Here virus and host must enter into a long-term interaction where selection is operating to a significant degree at partnership (in symbiological parlance, holobiontic) level. The virus has already culled the species genotype of the grey squirrel to one that allows longterm coexistence of virus and host, so it causes no overt disease while yet multiplying freely in the species. The extinction of the red squirrel would not cause the virus itself to become extinct. On the contrary the virus would benefit from the success of its partner, the grey squirrel, taking over the entire ecological niche of the red squirrel. This is one aspect of an evolutionary dynamic of plague viruses I termed "aggressive symbiosis".
In AIDS, even at the very peak of the present pandemic, and where HIV-1 and the human species are interacting mainly at selfish individual (or gene) level, there is the beginnings of such an aggressive symbiotic interaction, with virus and host influencing one another's evolution. This interaction, which is evolutionarily intense, involves the human HLA-B antigens -- coded by our Major Histocompatibility Complex (xMHC), which governs both our immune respones and our determination of self. (see Kiepiela P, et al, 2004. Nature 432: 769-74.). By the time the AIDS pandemic settles, it will have significantly changed the human species gene pool in populations where the epidemic has infected a large proportion of the population. What was formerly a small minority of people within those populations who were genetically more adapted to live with the persistent presence of the virus will survive and multiply, thus greatly increasing their proportion within the host population. In these the virus will be able to reproduce and spread without causing serious disease, an essential step to viral persistence. Moreover, since lentiviruses are capable of invading the germ line of mammals, and primates, HIV-1 could potentially enter into a long-term holobiontic union with the human genome, as have very large numbers of retroviruses with our human and pre-human ancestors.
This would take it to a new and powerful symbiogenetic stage where selection, operating at the holobiontic level of combined vertebrate and viral inheritance within the genome, would open up new evolutionary possibilities in the future.