My Salesforce colleagues just pointed out that the JDBC regression tests fail against 9.4beta3, because those tests include a check on the timezone Europe/Helsinki in 1920, and the latest IANA timezone database changed what they think the GMT offset was there/then.
This doubtless explains why buildfarm member piapiac has been unhappy since the TZ updates went in, though it sure doesn't provide much visibility into the exact failure :-(
I'd counsel following the precedent of commit 66b737cd9a72e10df4de1867ae7b675bc48d478c if you want to test any fractional-second GMT offsets: use Paris, whose location was pretty well established even back then.