So for example check modified, inode number against the previous version. If they are all a match consider skipping the checksum. This way you can skip almost all checksums on an incremental transfer. You can have false-negatives but this can often be mitigated by occasionally doing a full rescan or similar.

To be honest, I didn't even try this because the destination machine is Windows, I didnt think inode number had an equivalent in windows, but after googling, there might be a way.

Another great suggestion.

As for the full checksum, I went with a chunk size of the first and the last of the file, (the changes are very small, property based) but I don't think its as reliable.