Okay, I retract my argument.
When you said that, I decided I too would run a test (because I've seen wildly differing and vague numbers in public tests I've seen). So, I just did two tests. I took a completely random 500 file sampling of my library (a mix of sources from audiobooks to classical to jazz to ABBA), and converted them to FLAC three times. Once to v6 (the default in MC) to start, and then from that "source" to FLAC -0 and FLAC -8.
I did this twice, with two sets of files.
On the first try, the resulting -8 files were 12.6% smaller than the -0 files.
On the second try, they were 18.4% smaller.
That's not totally insignificant with a large library, and was (even in the small case) a bit higher than I'd expected (which was between 4-9%), so I retract my argument. They're worth leaving in for a 10-20% difference, since it is supported already anyway.
Just relabel them.