I thought about it and I think I'm asking the wrong question.
Scenario: Reading large NTFS compressed text logs versus uncompressed to be processed for web analytics
Assuming NTFS compressed files are compressed down to 10%, disk activity (assuming enough memory) should be 10% of what it would be versus reading an uncompressed file. Disk is the slowest part of the equation right? Yes, it takes some CPU time to uncompress the file but CPUs are fast enough so impact on CPU should be minimal. But the file still needs to go to the CPU so in effect, we're going back to a PIO-like mode rather than a DMA-like mode. Is there a performance impact there in getting the file read? Like I said, assume the CPU isn't being a bottleneck but it's part of the pathway in uncompressing the file.