I totally agree with this as now we get 50% of the builds failed even when a single line of code is modified, just because of the rounding errors.
The situation is even worse because exporting status via cc.xml files (CCTray) supports only success and failed, and the unstable is mapped as a failure without any way of changing this. So we see lots of "red" builds just because the code coverage went from 99.99% to 99.98%. Clearly the threshold for measuring a decrease should be a percent, not a "rounding error".
Maybe an option to set build unstable if coverage decreases by a certain percentage or value compared to previous build (ex: if code coverage drops by 2% from previous build, set build to unstable).