Read it again, carefully. It reads like advertising hype, which it is. Maybe they calibrate one unit in a production run, or something like that, but if their "enhanced manufacturing process" (which they don't explain) ensures uniformity from unit to unit, then they are all bad!!!??
If you want assurance of uniformity, you should receive a calibration certificate specific to serial number, dated, and signed, with the tolerances and test results. Then, there should be some way to recalibrate it when it drifts out of calibration. That's the way it is done with any other kind of calibrated tool or measuring device. I told Epson this, and they agreed and wrote it down as a suggestion. I told them about Epson's own ColorBase program, only available from the European Epson web sites, or the UK web site. Guess what... They hadn't heard of it.
An Epson service technician can indeed calibrate the machine to the hardware level inside the printer. Or, if it works for you, there is the ColorBase software I mentioned above that you can run for each of the paper types you use (see Eric Chan's website), but you need an Eye-one spectrometer. Unfortunately, we have no Epson tech reps in Hawaii, and the ColorBase program reported that the mismatch was too severe for it to properly calibrate my unit, so I had to resort to other methods.
My 3800 was producing muddy color out of the box, as was the one working replacement. The Advanced Black and White mode worked fine, and is very linear, but the color was a horrible mess. It looked like I was double profiling everything, but I wasn't.
As for a method of getting results that match the prints from two machines, you can get very close with a method that Barry Haynes, Wendy Crumpler, and Sean Duggan have in chapter 16 of their book, "Photoshop Artistry for Photographers Using Photoshop CS2 and Beyond." It's a great book that I highly recommend.
In essence, you make a copy image look like the print you made using adjustment layers, then you adjust it again to match the original screen image, then you shut off the original adjustment layers and print from that, thus compensating for the mismatch. You wind up with a recorded action, then, that you play when you want to match the output of the other machine. It's sort of a poor man's profile editor without a spectrometer.
Aloha,
Aaron