in order to answer question(1) let me tell you that we have worked with Sequenom, SNPlex, Illumina, and Affymetrix (listed by throughput) during these past years, and what I've always heard from my lab colleagues was that Sequenom design was very flexible when dealing with very close SNPs. are you considering Illumina because the number of SNPs you want to test is ~768 or for any other reason? because if this number is relatively low I would recommend you investing some time doing some numbers, because even if Sequenom allows typing 384 samples at a time per plate, the highest multiplex allowed is ~40 SNPs, so if you have several thousands of samples you will have to calculate the number of plates to process all your genotypes, because that will give you an idea at least of how long it will take:
numberOfGenotypes = numberOfSamples * numberOfSNPs
numberOfPlates = numberOfGenotypes / ( 384 * 40 )
once you have this figure, if it suits your research, you may then ask for prices. considering the size of the project the cost per genotype will be really low (and accurate), but you'll have to multiply it by numberOfGenotypes and then take the final decision whether to use or not Sequenom. we no longer work with SNPlex, and I understand that Affy would be too extense for your needs, but if you need further advice regarding Sequenom or Affy just drop a line.
to answer question (2) I would have to refer to the standard way of selecting SNPs for an assay, which is to deal with tagSNPs when having problems to type certain SNPs. if you haven't come across this option yet, the idea would be to choose SNPs from the same LD block as the problematic SNPs because they segregate together as a single haplotype. unfortunately, the problem you may have to face is that LD variation among blocks of 60bp may not be significant.