This is a very interesting paper about detecting advanced type-3 clones, i.e. clones which are modifications of the original code, not just copy-paste or parameter difference.
They've studied a number of projects and came to a conclusion that modern tools can detect only 25% of the clones of type-3. the remaining 75% of the clones detected were false positives.
Depending on the algorithm used to detect the clones detected have different characteristics, e.g. length or complexity of modification.
Looking forward for more work along these lines.