This topic has been discussed extensively on FabFilter's forums, and after investigation, their engineers have determined the bug is on Apple's end, so I'm bringing the topic over here in hopes that, after some number of years, it'll finally be resolved. In summary, the bug is that Apple does not accurately compensate for the correct number of latency samples reported by FabFilter. This has been confirmed on most, if not all, OS's, and at least the last 13 months of LPX updates. Also, when bouncing in place with a FabFilter plugin (most noticeably in linear phase mode), the resulting audio is shifted forward in time very significantly. I'll copy the text of one of those posts here. To read this thread in full, including responses from FabFilter's engineers, see here https://www.fabfilter.com/forum/3302/logic-pro-x-latency-not-reporting
I've been running into an issue with latency not accurately reporting in LPX with all FabFilter plugins. Of course, I have latency compensation set to "all" in LPX settings, and this seems to be an issue only with FabFilter plugins. This has been making any parallel processing or bus processing useless in a mixing environment, and so far I've wasted hours trying to troubleshoot.
Here's an example in action. www.dropbox.com/s/ez6pqg4mu56uox7/lpx-latency.mov?dl=0
You can hear the LFO tool triggering the sidechain at a different point depending on which FabFilter plugins used. This is even the case when Pro-MB is set to Dynamic Phase and Pro-L2 is set to "transparent" with True Peak Limiting off, though the latency is not as bad. Regardless, LPX is not compensating for it. Also, for the issue to be corrected, the plugins must be removed from the channels completely – simply bypassing won't do the trick.