Silent data corruption applies more to BCLK overclocking (where you're also altering the PCI-E / SATA bus speeds). Changing just the multiplier, leaving BCLK at 100 and only doing a mild OC to 3.8-4.0ghz drastically reduces the chances of it happening. If it passes Prime / IBT, etc, there's likely to be zero "silent errors" unless you've got a faulty CPU / motherboard / RAM (in which case you'll have problems at any clock speed).
I was thinking the same thing. Silent data corruption is going to involve the storage subsystem unless your CPU + memory controller are routinely doing bad read/writes from/to memory. That's going to show up as unstable system behavior. Sure, you CAN get data corruption on such an unstable system, but the signs will be there. Been there, done that.
If your system isn't crashing and it is passing stuff like Memtest86+ and linpack (max memory) then, if there are any potential bad writes possible, they won't be due to the CPU/mem controller no matter what the overclock. The idea that a G3258 @ 3.8 ghz + stock vcore is going to hose up a bunch of spreadsheets seems more reactionary than based in fact, especially if the chip has run Prime95 or linpack for 24+ hours.
Overclocking for the office just seems like a terrible idea when you have any significant number of machines to set up, run, and maintain. If it's a soho setup with one computer, maybe, but any IT department of serious size is not going to want to tune hundreds of computers for a little speed bump. Overclocking (even mildly) will also increase the chance that the system(s) will be sensitive to dust buildup over time, so they'll require maintenance in the form of cleaning that might not be required for stock builds. Not a big problem for one machine, but for more than 2-3, it's headache city.