I am not absolutely sure if I understood your data structure correctly, by I would go for a solution with "a table where each record corresponds to a SNP and an individual". Even though such a table will become large, Postgres should be able to handle this. In particular, if you set up unique index over the SNP and individual. Also, this format provides the most flexibility in terms of querying and extending the database (e.g., which individuals have a particular SNP).
I strongly vote against using array data types. Updates of arrays will become a pain if you need to add another individual. Additionally, you have to manage the individual-to-array-index by yourself. Postgres is able to ensure referential integrity if you store data in long-table format.
Actually, I use the long-table format to store micro-array expression data in a PostgreSQL database (and other kinds of data, but for now mirco-array is the largest data set). For example, the following selection queries the normalized intensities of all probes measuring EGFR expression in all samples. The query requires less than <300ms for the first execution and ~25ms for subsequent executions (including changing the gene name)
select ds.name, probeset.probeset, symbol.accession, gcrma.gcrma
from bioinfo_hgu133_gcrmas as gcrma, bioinfo_hgu133_probesets as probeset, bioinfo_hgu133_mappings as mapping, bioinfo_genomic_genesymbols as symbol, bioinfo_hgu133_datasets as dataset, bioinfo_datasets as ds
where gxp.probeset_id = probeset.id
and probeset.id = mapping.probeset_id
and mapping.genesymbol_id = symbol.id
and gxp.hgu133_dataset_id = dataset.id
and dataset.dataset_id = ds.id
and symbol.accession = 'EGFR'
The query runs on 792 individual datasets (i.e., micro arrays), >50k probesets per dataset mapped to 90k genes. The total number of individual measurements (gcrma.gcrma) is 38 million.
