I tried jminer. It's 15 seconds to read my 25TB of plots on my GPU comparet to 42 seconds with Blago's CPU miner. The jminer, however, have the same bug in Windows which the Blago miner had some versions ago - it consumes the RAM (windows disk catching) and you can't use your PC for anything else (heavy reading from pagefile). This was fixed in the Blago's miner, so a fix is possible.
Guess you are running Win7, it is the only OS having theese disk caching problems, you could solve them by using Win8+.
jminer is running in JavaVM using java.nio (http://docs.oracle.com/javase/8/docs/api/java/nio/package-summary.html) for file operations.
(https://github.com/de-luxe/burstcoin-jminer/blob/master/src/main/java/burstcoin/jminer/core/reader/task/ReaderLoadDriveTask.java)
Sorry, but there will be no native Win7 special cases implemented by me, even if it would be possible (however) ...
From my point of view, this is a Win7 issue ... and not a bug in jminer.
Thank you for your answer. Is the software cache needed in this case at all? After all you are reading different data each block. As far as I understand Blago's answer it's setting a single flag for all Windows versions:
@luxe,
https://github.com/nyholku/purejavacomm/blob/master/src/jtermios/windows/WinAPI.java
http://www.docjar.com/html/api/sun/nio/fs/WindowsConstants.java.html
Upd: