[From page 521 of "The Cost of Natural Selection." All comments in square brackets and emphasis are mine.]
We do not know how many loci two 'good' but fairly closely related species differ. Their taxonomic characters may depend mainly on as few as twenty gene substitutions. But there is every reason to think that substitutions have occurred at a great many other loci. [Haldane continues the paragraph with a discussion of the viabilities of 18 recessive mutations studied by Punnett in Lathyrus odoratus and how the viability of these mutations seemed to be increasing with time (the older mutations invariably had higher viabilities than the newer mutations). Haldane felt that this was due to the accumulation of modifiers (genes at other loci)  that improved the viability of the 18 visible mutations. This was Haldane's justification for claiming that even though twenty mutations might explain the taxonomic character differences between two species, substitutions may have occurred at a great many other loci.]

[In the following paragraph, Haldane begins...]
Presumably the same kind of process occurs in evolution. The number of loci in a vertebrate species has been estimated at about 40,000. 'Good' species, even when closely related, may differ at several thousand loci, even if the differences at most of them are very slight. But it takes as many deaths, or their equivalents, to replace a gene by one producing a barely distinguishable phenotype as by one producing a very different one. If two species differ at 1000 loci, and the mean rate of gene substitution, as has been suggested, is one per 300 generations, it will take 300,000 generations to generate an interspecific difference. It may take a good deal more, for if an allele a1 is replaced by a10, the population may pass through stages where the commonest genotype is a1a1, a2a2, a3a3, and so on, successively, the various alleles in turn giving maximal fitness in the existing existing environment and the residual environment.  Simpson (1953) finds the mean life of a genus of Carnivora to be about 8 million years. That of a species in horotelic vertebrate evolution may average about a million.

Zeuner (1945), after a very full discussion of the Pleistocene fossil record, concluded that in mammals, about 500,000 years were required for the evolution of a new species, though in the mole genera Mimomys and Arvicola the rate was somewhat greater. Some insects seem to have evolved at about the same rate, while other insects, and all molluscs, evolved more slowly. He estimated the total duration of the Pleistocene at 600,000 years, but some later authors would about halve that figure [note that these assumptions turned out to be quite wrong - the duration of the Pleistocene has been firmly established at over 1.5 million years in the years since Haldane wrote this]. On the other hand, changes during the Pleistocene were unusually rapid, and evolution, therefor, also probably unusually rapid. The agreement with the theory here developed is satisfactory.

[To me, this last sentence seems to indicate that Haldane saw no problem with his estimated rate of 1 substitution per 300 generations and the rate of evolution inferred from the fossil record. Additionally, Haldane seems satisfied that one to two thousand substitutions would be sufficient to differentiate two close species. Since 1,667 (the approximate number of substitutional differences expected between humans and chimpanzees if it has been 5 million years since they shared a common ancestor) substitutions falls well within this range, it appears that Haldane would have seen no problem with this number of substitutional differences between humans and chimpanzees. His arguments concerning multiple substitutions at the same locus seem weak since he is talking about substitutions at 1,667 of 40,000 loci (that works out to about 1 substitution per 24 loci) , there is no reason to expect that there would be a great deal of multiple substitutions at the same locus. At any rate, Haldane made no effort to quantify either the number of serial substitutions that would be expected at a single locus nor the number of modifiers that would be expected to raise the number of genes required to explain the taxonomic differences between to good but fairly closely related species from twenty all the way up to one to two thousand.]