[BioC] number of Replicates - Fold change vs Limma or Siggenes/SAM

[Luckey, John]

I am looking to identify differentially expressed genes from five different cell types with five, five, four, four, and three affymetrix replicates each. (I will be using gcrma as preprocessing step)
 
I seem to remember a paper that showed that T-tests did not predict known spiked-in changes better than simple fold change until one had 8 or more replicates.
 
Is there data with known spiked in datasets (or a consensus) that limma or sam/siggenes performs better than straight fold change with such low replicate analysis?
 
Sincerely,
John Luckey