PICRUSt question- whether to normalize gene table

I am using PICRUSt (through the Nephele platform). From Langille et a. (2013), the OTU table is normalized, the metagenome is inferred, and the gene table is calculated. My colleagues and I are unsure whether we should normalize the counts in the gene table. It seems that many researchers do so- with our own data, the results make more sense when the gene table is normalized. Using gene counts (without normalization), the data are highly skewed. Your advice is appreciated.

I would recommend normalizing the gene counts.