Why do you say there'll be a performance loss on commit if the log is getting cleared (of inactive portions only) on a checkpoint?
Got some tests that show the performance degradation?
Looking at it rationally rather than empirically, I'd say that since a log file is written to sequentially, any activity (such as truncation) that moves the disk heads away from the end of the file is likely to have some impact on performance. I haven't done any tests to ascertain how noticeable such an impact is.
John