No, Dawkins doesn't talk about selfish instincts. He talks about genes, and the issue here is the possibility of selfish genes determining altruistic instincts. (Or cooperation-inducing instincts, or even self-harmful instincts.) Dawkins agues a great deal about potential conflict between the interests of individuals and the 'interests' of their genes. You really should read the book.

parallel to straight socio-Darwinistic arguments. In particular, his "selfish gene" metaphor super-emphisises the role of an individual - nothing matters but survival of the smallest subject.

How is that parallel to Social Darwinism? It is obviously reductionist, in which direction your critisism runs. But the problem with Social Darwinism is not making reductionist arguments (and that's not what they do), but confusing description (what happens and why) and desirability (what should happen), treating nature as some mythical force rather than changing and replaced by our artifical environment & culture, having no clue about the actual human genetic variance but making simplistic (racist) assumptions about it, and having a very narrow (again ideological advocacy parading as science) view of what can constitute 'fitness'.

Dawkins kind of denies meaningful selection on higher than genetic levels

What I'd accuse Dawkins of is denying the importance of evolutionary processes other than adaptive selection. In his later books and articles, when he confronts more scientific criticisms, that's pretty apparent -- he acknowledges things like genetic drift and macroevolutionary processes, but treats them as uninteresting detail when compared to the order-creating(/retaining/accumulating) selection. Dawkins's opponents ascribe this to his reductionism.

Much more meaningful selection does happen on organism level, and sometimes - on group level, when cooperation is a straight necessity (though not necessarily sufficiency) for survival, however you spin that.

...and you accuse others of sounding like Social Darwinists?...

First, you read the Extended Phenotype, but still don't appear to make a difference between genes and phenotype. Selection considered by Dawkins (or Darwin) always acts on the phenotype, be it an organism, a wolf pack or a termite mound. A phenotype is always the result of multiple genes acting in concert (sometimes genes of multiple species, as you say elsewhere). Question is, was there a genetic difference that made a difference? If there was no genetic difference, no biological evolution was involved. If there was difference but it made no difference, we have a case of genetic drift, not selection. (One of the things Dawkins recognises but thinks uninteresting.)

Now, if there was genetic difference and that mattered, how extensive is it? Surely not the entire range of cooperating genes in an organism: there isn't even difference in most. And by all likelyhood, not even the entire genetic variation present in the competing sides (be them organisms or groups of organisms), especially if for genes with variation within, not just between groups.

Viewing group-level conflicts as competition of gene alliances comes dangerously close to racism.

In particular, what is "ultimate selfishness"?

I'm not sure what you refer to. Could you give a quote?

_{*Lunatic*, n.
One whose delusions are out of fashion.}