From what I've read, it looks like there is a way for a program with 'user' level permissions to figure out the memory layout of 'system' level processes when both are present in the CPU cache, and then use that as the starting point for another exploit. If you have to flush the CPU cache when switching between 'user' and 'system' permissioned processes, that's going to slow things down a bit, but it's also going to vary a lot with the particular situation. I think it's too hard at this stage to say exactly how it's going to impact usability, but I sincerely doubt it's going to be a big issue for most use cases.
The article from this morning gave the 5-30% performance reduction figure. However, here is an updated statement:
' “Intel has begun providing software and firmware updates to mitigate these exploits,” Intel said in a statement. “Contrary to some reports, any performance impacts are workload-dependent, and, for the average computer user, should not be significant and will be mitigated over time.” '
So, you're right. It's easy for the media to misinterpret numbers and run away with their conclusions. Based on what you're saying about the exact mechanism behind the bug, it makes sense that potential slowdowns will be highly dependent on workload. For scientific calculations running on e.g. BOINC, I wonder if flushing the CPU cache between 'user' and 'system' permissioned processes would even really matter. This is def going beyond my area of expertise. I'm going to track with the developing news though, and hopefully we eventually get a clearer picture of what the impact actually is.