I have MTU 1508 set up but I couldn’t measure a 0.7% improvement like that, because the noise in speed testers is far too great. I routinely get 10% variation in some speed testers sometimes, but other times they’re more reproducible. If TCP is used - which it shouldn’t be - then so much depends on the behaviour of TCP software at both ends, not in the link.
The letmecheck.it tester does a good job of testing for MTU, but watch out, because - as has been mentioned in earlier threads - their stupid setup causes things to fail if you are able to use IPv6 at your end.