And just to follow up...
 

I sent our F350 in to Teaneck NJ for hot pixels that would not reset using the black balance method. The repair description states that "RPN management performed." So it looks like electronic management was used to mask/blend the offending pixels and no block replacement was performed.

I must give kudos to Sony for the turnaround time, though. I shipped on Tuesday and had it back on Friday. Can't complain too much about that.

Don't get me wrong. I don't believe that it is acceptable to have hot or dead pixels spoiling your pictures at 0db gain within the first year of ownership of an expensive piece of kit.

I wish I knew why this seems to be a problem that affects high end cameras more than cheap consumer cameras.

I don't have a problem with pixel masking or re-mapping as long as it is transparent in use, which in most cases it appears to be. RPM (residual pixel noise) masking improves all the time. Older digibetas could only mask about 32 pixels, newer cams many more. AFAIK Sony have 2 layers of RPN mapping. One that is done using the Black Balance method which has a limited amount of memory and then a second bank that can be programmed by the factory or repair center. When you send a camera in for RPN work they copy the data from the user memory to the hard memory so that the user memory can then mask a further batch of pixels. At some point however you run out of memory and then you may be looking at a new block.

That's a bit different than the explanation I got while training at Sony last year.

Black balancing does nothing to mask stuck pixels. Black balancing observes the output from each pixel site in a total blackness scenario, and zeros the gain for any pixel still producing residual current. This will not resolve a pixel that is truly stuck on (defective). Black balancing gets rid of residual thermal noise where the pixel should be producing zero output current. Thermal noise would be incorrectly interpreted as image data.

RPN is done in the service menus and has both a normal memory and a separate memory for slow shutter RPN data. Pixel masking shuts off (or ignores) the stuck pixel site and uses nearest neighbor interpolation to approximate the disabled pixel's data. The interpolation will work fine up to a point. Too many approximations will start to noticeably degrade resolution. I like that it has a dedicated RPN routine for slow shutter since many of those pixels aren't really stuck on and will be just fine under normal operation and shouldn't be shut off unnecessarily. They just have a tendency to ramp up too much if allowed to accumulate a charge for too long. Almost like a thermal runaway condition.

-gb-