The first answer you got suggested using a database. Databases are designed to adapt to larger-than-memory datasets. Why not use a database? (e.g. http://www.postgresql.org/).

(I'm also a little curious what data you are dealing with: sequencing, microarrays, snp's? .cab files? Does that mean, windows? Does that mean desktop environment (as opposed to $BigServer?) Affymetrix? Anyway, it may not be essential to know but it may make it more interesting.. )