Hello Fred,
Yep, I mean reducing the training data set using --p-n-reads-learn.
To speed this process up, you might consider renting a super-computer from Amazon and running qiime on it. This process is a little technical the first time you set it up, but can remove a major bottleneck in your analysis.
Seriously, have you considered renting a super-computer? I've highlighted three good ones on this list, and they cost under $2 a hour. For the price of running PCR on one sample, you could do all this analysis and get some results. Amazon EC2 Instance Comparison
Colin