To download microarray data, the first stop is GEO Datasets (http://www.ncbi.nlm.nih.gov/gds). You can search by tumor type, organism, platform, pubmed ID etc. Anyone publishing microarray data has to make it publically available, and that is the most popular place. There is also the ArrayExpress website for those who prefer the EBI (http://www.ebi.ac.uk/arrayexpress/), though I tend to find ArrayExpress to be a subset (albeit a slicker looking subset) of GEO. You can browse the normalized data directly on some of the GEO datasets without even downloading, though it's straightforward to pull the data directly from their FTP site or via various tools such as R.
You can also try to correlate these expressions to the Amplification data provided by the CONAN tool available at the Sanger website.
One of the most authoritative resources for this is CGAP and CGCI from the (US) National Cancer Institute. The NCI's Cancer Genome Anatomy Project sought to determine the gene expression profiles of normal, precancer, and cancer cells, leading eventually to improved detection, diagnosis, and treatment for the patient. Resources generated by the CGAP initiative are available to the broad cancer community. Interconnected modules provide access to all CGAP data, bioinformatic analysis tools, and biological resources allowing the user to find "in silico" answers to biological questions in a fraction of the time it once took in the laboratory.
There are several links to all types of data and tools: genes, chromosomes, tissues, SAGE Genie, RNAi, pathways and tools.
This should provide you with a lot of data with a specific cancer focus.
Genevestigator contains a wide compendium of gene expression in cancer. It recently got mapped to the ICD-10 and ICD-O-3 classification. User friendly tools allow you to compare expression across various cancer types and compare with normal, non-tumor tissues.