Given the rather specific nature of your situation, I would suggest contacting customer support at Affymetrix. While a few of the BioStar members may be able to comment on their experience with genotyping tumor vs normal, the Affy support desk will have information from many more customers as well as their own in-house development teams.

Firstly the standard way to estimate tumor percentage from histological slides usually under estimates the number of non-tumor cells in a sample. Tumor cells are often larger than stromal cells or tumor infiltrating lymfocytes, while they contain the same amount of DNA. There is generally more dilution than you think there is.

Secondly genotyping algorithms are geared towards classifying bi-allelic loci. The genetic changes in cancer violate the assumptions in these algorithms.

It is also useful to check in which direction the genotype shifts. Ideally you expect that only heterozygous SNPs in the matched normal become homozygous in the tumor. SNPs that change from a homozygous state in the normal sample tell something about the genotyping error rate.

Also the genetic changes in cancer often have sizes in the order of megabases or larger. You expect SNPs with LOH to be clustered.

Not all cancers have the same amount of genetic changes. Sarcomas and lymphomas can have very widespread changes, some carcinomas in patients with a hereditary cancer gene can have very circumscript changes.