It seems that additional information should be captured and reported, including:

- Some approximation of the manufacture date (performance of the same model could change over time due to manufacturing process and/or supplier changes, etc.), either from a serial number and/or purchase date, etc.

- More formal statistics (stddev, etc.) on collections of individual results to monitor consistency and flag inconclusive/suspicious results. You mention that you discard outliers, but in some cases outliers might reveal actual inconsistencies in product quality

Since this data will certainly influence buying decisions, there needs to be some protection against data manipulation. Can your benchmark utility generate a signed binary result file that could be submitted? Do flash drives report some sort of unalterable id to identify their manufacturer, model, etc. that could be embedded in the result file?

Without additional info and safeguards in place, there's a high risk here of publishing even unintentional misleading results. In the end the best you can do is document your collection methods and disclaim absolute validity of the results, allowing readers make their best informed decision.

I bring up these issues constructively and hope they don't deter you from providing a really valuable source of information...

Should be some interesting results once USB 3 hits the mainstream with the rest of the e-creatures.